qubes-doc/basics_dev/gsoc.md

layout	title	permalink	redirect_from
sidebar	Google Summer of Code	/gsoc/	/GSoC/

2017 Google Summer of Code

Information for Students

Thank you for your interest in participating in the Google Summer of Code program with the Qubes OS team. You can read more about the Google Summer of Code program at the official website and the official FAQ.

Being accepted as a Google Summer of Code student is quite competitive. Students wishing to participate in the Summer of Code must be aware that you will be required to produce code for Qubes OS for 3 months. Your mentors, Qubes developers, will dedicate a portion of their time towards mentoring you. Therefore, we seek candidates who are committed to helping Qubes long-term and are willing to do quality work and be proactive in communicating with your mentor.

You don't have to be a proven developer -- in fact, this whole program is meant to facilitate joining Qubes and other free and open source communities. The Qubes community maintains information about contributing to Qubes development and how to send patches. In order to contribute code to the Qubes project, you must be able to sign your code.

You should start learning the components that you plan on working on before the start date. Qubes developers are available on the mailing lists for help. The GSoC timeline reserves a lot of time for bonding with the project -- use that time wisely. Good communication is key, you should plan to communicate with your team daily and formally report progress and plans weekly. Students who neglect active communication will be failed.

Overview of Steps

• Join the qubes-devel list and introduce yourself, and meet your fellow developers
• Read Google's instructions for participating and the GSoC Student Manual
• Take a look at the list of ideas below
• Come up with a project that you are interested in (and feel free to propose your own! Don't feel limited by the list below.)
• Read the Student Proposal guidelines below
• Write a first draft proposal and send it to the qubes-devel mailing list for review
• Submit proposal using Google's web interface ahead of the deadline (this requires a Google Account!)
• Submit proof of enrollment well ahead of the deadline

Coming up with an interesting idea that you can realistically achieve in the time available to you (one summer) is probably the most difficult part. We strongly recommend getting involved in advance of the beginning of GSoC, and we will look favorably on applications from students who have already started to act like free and open source developers.

Before the summer starts, there are some preparatory tasks which are highly encouraged. First, if you aren't already, definitely start using Qubes as your primary OS as soon as possible! Also, it is encouraged that you become familiar and comfortable with the Qubes development workflow sooner than later. A good way to do this (and also a great way to stand out as an awesome applicant and make us want to accept you!) might be to pick up some issues from qubes-issues (our issue-tracking repo) and submit some patches addressing them. Some suitable issues might be those with tags "help wanted" and "P: minor" (although more significant things are also welcome, of course). Doing this will get you some practice with qubes-builder, our code-signing policies, and some familiarity with our code base in general so you are ready to hit the ground running come summer.

Student proposal guidelines

A project proposal is what you will be judged upon. Write a clear proposal on what you plan to do, the scope of your project, and why we should choose you to do it. Proposals are the basis of the GSoC projects and therefore one of the most important things to do well. The proposal is not only the basis of our decision of which student to choose, it has also an effect on Google's decision as to how many student slots are assigned to Qubes.

Below is the application template:

# Introduction

Every software project should solve a problem. Before offering the solution (your Google Summer of Code project), you should first define the problem. What’s the current state of things? What’s the issue you wish to solve and why? Then you should conclude with a sentence or two about your solution. Include links to discussions, features, or bugs that describe the problem further if necessary.

# Project goals

Be short and to the point, and perhaps format it as a list. Propose a clear list of deliverables, explaining exactly what you promise to do and what you do not plan to do. “Future developments” can be mentioned, but your promise for the Google Summer of Code term is what counts.

# Implementation

Be detailed. Describe what you plan to do as a solution for the problem you defined above. Include technical details, showing that you understand the technology. Illustrate key technical elements of your proposed solution in reasonable detail.

# Timeline

Show that you understand the problem, have a solution, have also broken it down into manageable parts, and that you have a realistic plan on how to accomplish your goal. Here you set expectations, so don’t make promises you can’t keep. A modest, realistic and detailed timeline is better than promising the impossible.

If you have other commitments during GSoC, such as a job, vacation, exams, internship, seminars, or papers to write, disclose them here. GSoC should be treated like a full-time job, and we will expect approximately 40 hours of work per week. If you have conflicts, explain how you will work around them. If you are found to have conflicts which you did not disclose, you may be failed.

Open and clear communication is of utmost importance. Include your plans for communication in your proposal; daily if possible. You will need to initiate weekly formal communications such as a detailed email to the qubes-devel mailing list. Lack of communication will result in you being failed.

# About me

Provide your contact information and write a few sentences about you and why you think you are the best for this job. Prior contributions to Qubes are helpful; list your commits. Name people (other developers, students, professors) who can act as a reference for you. Mention your field of study if necessary. Now is the time to join the relevant mailing lists. We want you to be a part of our community, not just contribute your code.

Tell us if you are submitting proposals to other organizations, and whether or not you would choose Qubes if given the choice.

Other things to think about:
* Are you comfortable working independently under a supervisor or mentor who is several thousand miles away, and perhaps 12 time zones away? How will you work with your mentor to track your work? Have you worked in this style before?
* If your native language is not English, are you comfortable working closely with a supervisor whose native language is English? What is your native language, as that may help us find a mentor who has the same native language?
* After you have written your proposal, you should get it reviewed. Do not rely on the Qubes mentors to do it for you via the web interface, although we will try to comment on every proposal. It is wise to ask a colleague or a developer to critique your proposal. Clarity and completeness are important.

Project Ideas

These project ideas were contributed by our developers and may be incomplete. If you are interested in submitting a proposal based on these ideas, you should contact the qubes-devel mailing list and associated GitHub issue to learn more about the idea.

### Adding a Proposal

**Project**: Something that you're totally excited about

**Brief explanation**: What is the project, where does the code live?

**Expected results**: What is the expected result in the timeframe given

**Knowledge prerequisite**: Pre-requisites for working on the project. What coding language and knowledge is needed? 
If applicable, links to more information or discussions

**Mentor**: Name and email address.

Qubes MIME handlers

Project: Qubes MIME handlers

Brief explanation: #441 (including remembering decision whether some file should be opened in DispVM or locally)

Expected results:

• Design mechanism for recognising which files should be opened locally and which in Disposable VM. This mechanism should:
  • Respect default action like "by default open files in Disposable VM" (this may be about files downloaded from the internet, transferred from other VM etc).
  • Allow setting persistent flag for a file that should be opened in specific way ("locally"); this flag should local to the VM - it shouldn't be possible to preserve (or even fabricate) the flag while transferring the file from/to VM.
  • See linked ticket for simple ideas.
• Implement generic file handler to apply this mechanism; it should work regardless of file type, and if file is chosen to be opened locally, normal (XDG) rules of choosing application should apply.
• Setting/unsetting the flag should be easy - like if once file is chosen to be opened locally, it should remember that decision.
• Preferably use generic mechanism to integrate it into file managers (XDG standards). If not possible - integrate with Nautilus and Dolphin.
• Optionally implement the same for Windows.
• Document the mechanism (how the flag is stored, how mechanism is plugged into file managers etc).
• Write unit tests and integration tests.

Knowledge prerequisite:

• XDG standards
• Bash or Python scripting
• Basic knowledge of configuration/extension for file managers

Mentor: Marek Marczykowski-Górecki

Template manager, new template distribution mechanism

Project: Template manager, new template distribution mechanism

Brief explanation: Template VMs currently are distributed using RPM packages. There are multiple problems with that, mostly related to static nature of RPM package (what files belong to the package). This means such Template VM cannot be renamed, migrated to another storage (like LVM), etc. Also we don't want RPM to automatically update template package itself (which would override all the user changes there). More details: #2064, #2534.

Expected results:

• Design new mechanism for distributing templates (possibly including some package format - either reuse something already existing, or design new one). The mechanism needs to handle:
  • integrity protection (digital signatures), not parsing any data in dom0 prior to signature verification
  • efficient handling of large sparse files
  • ability to deploy the template into various storage mechanisms (sparse files, LVM thin volumes etc).
  • template metadata, templates repository - enable the user to browse available templates (probably should be done in dedicated VM, or Disposable VM)
• Implement the above mechanism:
  • tool to download named template - should perform download operation in some VM (as dom0 have no network access), then transfer the data to dom0, verify its integrity and then create Template VM and feed it's root filesystem image with downloaded data.
  • tool to browse templates repository - both CLI and GUI (preferably in (py)GTK)
  • integrate both tools - user should be able to choose some template to be installed from repository browsing tool - see #1705 for some idea (this one lack integrity verification, but similar service could be developed with that added)
• If new "package" format is developed, add support for it into linux-template-builder.
• Document the mechanism.
• Write unit tests and integration tests.

Knowledge prerequisite:

• Large files (disk images) handling (sparse files, archive formats)
• Bash and Python scripting
• Data integrity handling - digital signatures (gpg2, gpgv2)
• PyGTK
• RPM package format, (yum) repository basics

Mentor: Marek Marczykowski-Górecki

Qubes Live USB

Project: Revive Qubes Live USB, integrate it with installer

Brief explanation: Qubes Live USB is based on Fedora tools to build live distributions. But for Qubes we need some adjustments: starting Xen instead of Linux kernel, smarter copy-on-write handling (we run there multiple VMs, so a lot more data to save) and few more. Additionally in Qubes 3.2 we have so many default VMs that default installation does not fit in 16GB image (default value) - some subset of those VMs should be chosen. Ideally we'd like to have just one image being both live system and installation image. More details: #1552, #1965.

Expected results:

• Adjust set of VMs and templates included in live edition.
• Update and fix build scripts for recent Qubes OS version.
• Update startup script to mount appropriate directories as either copy-on-write (device-mapper snapshot), or tmpfs.
• Optimize memory usage: should be possible to run sys-net, sys-firewall, and at least two more VMs on 4GB machine. This include minimizing writes to copy-on-write layer and tmpfs (disable logging etc).
• Research option to install the system from live image. If feasible add this option.

Knowledge prerequisite:

• System startup sequence: bootloaders (isolinux, syslinux, grub, UEFI), initramfs, systemd.
• Python and Bash scripting
• Filesystems and block devices: loop devices, device-mapper, tmpfs, overlayfs, sparse files.

Mentor: Marek Marczykowski-Górecki

Unikernel-based firewallvm with Qubes firewall settings support

Project: Unikernel based firewallvm with Qubes firewall settings support

Brief explanation: blog post, repo

Expected results: A firewall implemented as a unikernel which supports all the networking-related functionality as the default sys-firewall VM, including configuration via Qubes Manager. Other duties currently assigned to sys-firewall such as the update proxy may need to be appropriately migrated first.

Knowledge prerequisite:

• OCaml + MirageOS or other unikernel framework,
• Xen network stack,
• Qubes networking model & firewall semantics.

Mentor: Thomas Leonard, Marek Marczykowski-Górecki

IPv6 support

Project: IPv6 support

Brief explanation: Add support for native IPv6 in Qubes VMs. This should include IPv6 routing (+NAT...), IPv6-aware firewall, DNS configuration, dealing with IPv6 being available or not in directly connected network. See #718 for more details.

Expected results:

• Add IPv6 handling to network configuration scripts in VMs
• Add support for IPv6 in Qubes firewall (including CLI/GUI tools to configure it)
• Design and implement simple mechanism to propagate information about IPv6 being available at all (if necessary). This should be aware of ProxyVMs potentially adding/removing IPv6 support - like VPN, Tor etc.
• Add unit tests and integration tests for both configuration scripts and UI enhancements.
• Update documentation.

Knowledge prerequisite:

• network protocols, especially IPv6, TCP, DNS, DHCPv6, ICMPv6 (including autoconfiguration)
• ip(6)tables, nftables, NAT
• Python and Bash scripting
• network configuration on Linux: ip tool, configuration files on Debian and Fedora, NetworkManager

Mentor: Marek Marczykowski-Górecki

Thunderbird, Firefox and Chrome extensions

Project: additional Thunderbird, Firefox and Chrome extensions

Brief explanation:

• browser/mail: open link in vm
• browser/mail: open link in dispvm
• browser: save destination to vm
• mail: add whitelisted senders option (address-based and signing key-based) #845

Expected results:

• Extend existing Thunderbird extension to decide on action (where to open/save attachments) based on message sender - recognized as email address, or signing key
• Add Firefox extension to open links in Disposable VM / selected VM (right-click option and a default action for not-whitelisted URLs/domains)
• The same for Chrome
• Add tests for above enhancements
• Update user documentation

Knowledge prerequisite:

• writing Thunderbird/Firefox extensions (XUL, javascript)
• writing Chrome extensions (javascript)

Mentor: Jean-Philippe Ouellet

LogVM(s)

Project: LogVM(s)

Brief explanation: Qubes AppVMs do not have persistent /var (on purpose). It would be useful to send logs generated by various VMs to a dedicated log-collecting VM. This way logs will not only survive VM shutdown, but also be immune to altering past entries. See #830 for details.

Expected results:

• Design a simple protocol for transferring logs. The less metadata (parsed in log-collecting VM) the better.
• Implement log collecting service. Besides logs itself, should save information about logs origin (VM name) and timestamp. The service should not trust sending VM in any of those.
• Implement log forwarder compatible with systemd-journald and rsyslog. A mechanism (service/plugin) fetching logs in real time from those and sending to log-collecting VM over qrexec service.
• Document the protocol.
• Write unit tests and integration tests.

Knowledge prerequisite:

• syslog
• systemd
• Python/Bash scripting

Mentor: Jean-Philippe Ouellet

GUI improvements

Project: GUI improvements

Brief explanation:

• GUI for enabling USB keyboard: #2329
• GUI for enabling USB passthrough: #2328
• GUI interface for /etc/qubes/guid.conf: #2304
• Improving inter-VM file copy / move UX master ticket: #1839
• and comprehensive list of GUI issues: #1117

Expected results:

• Add/enhance GUI tools to configure/do things mentioned in description above. Reasonable subset of those things is acceptable.
• Write tests for added elements.

Knowledge prerequisite:

• Python, PyGTK

Mentor: Jean-Philippe Ouellet

Xen GPU pass-through for Intel integrated GPUs

Project: Xen GPU pass-through for Intel integrated GPUs (largely independent of Qubes)

Brief explanation: This project is prerequisite to full GUI domain support, where all desktop environment is running in dedicated VM, isolated from dom0. There is already some support for GPU passthrough in Xen, but needs to be integrated in to Qubes and probably really make working, even when using qemu in stubdomain. GUI domain should be a HVM domain (not PV). This should be done without compromising Qubes security features, especially: using VT-d for protection against DMA attacks, using stubdomain for sandboxing qemu process (if needed) - qemu running in dom0 is not acceptable. More details in #2618.

Expected results:

• Ability to start a VM with GPU connected. VM should be able to handle video output (both laptop internal display, and external monitors if apply). That VM also should be able to use hardware acceleration.
• This project may require patching any/all of Xen hypervisor, Libvirt, Qemu, Linux. In such a case, patches should be submitted to appropriate upstream project.
• It's ok to focus on a specific, relatively new Intel-based system with Intel integrated GPU.

Knowledge prerequisite:

• C language
• Kernel/hypervisor debugging
• Basics of x86_64 architecture, PCIe devices handling (DMA, MMIO, interrupts), IOMMU (aka VT-d)
• Xen hypervisor architecture

Mentor: Marek Marczykowski-Górecki

Whonix IPv6 and nftables support

Project: Whonix IPv6 and nftables support

Brief explanation: T509

Expected results:

• Work at upstream Tor: An older version of https://trac.torproject.org/projects/tor/wiki/doc/TransparentProxy page was the origin of Whonix. Update that page for nftables / IPv6 support without mentioning Whonix. Then discuss that on the tor-talk mailing list for wider input. - https://trac.torproject.org/projects/tor/ticket/21397
• implement corridor feature request add IPv6 support / port to nftables - https://github.com/rustybird/corridor/issues/39
• port whonix-gw-firewall to nftables
• port whonix-ws-firewall to nftables
• make connections to IPv6 Tor relays work
• make connections to IPv6 destinations work

Knowledge prerequisite:

Mentor: Patrick Schleizer

Standalone connection wizard for Tor pluggable transports

Project: Standalone connection wizard for Tor pluggable transports

Brief explanation: #1938, https://www.whonix.org/blog/connection-bridge-wizard, https://github.com/Whonix/anon-connection-wizard

Expected results:

Users are presented with a GUI where they can select different bridges to use to connect to Tor if it is censored in their country/region, just like with the Tor Browser.

Knowledge prerequisite:

Mentor: Patrick Schleizer

Leverage modern static & dynamic analysis

Project: Leverage modern static & dynamic analysis

Brief explanation:

Expected results: Stand up tooling to automatically run various tools against the Qubes code base, and address as many found issues as possible.

Knowledge prerequisite: Familiarity with various analysis tools & techniques, including but not limited to: valgrind, coverity, clang's sanitizers, guided fuzzing.

Mentor: Jean-Philippe Ouellet

Formally analyze how untrusted inputs propagate through the Qubes code base

Project: Formally analyze how untrusted inputs propagate through the Qubes code base

Brief explanation: It would be useful to have a rigorous understanding of what code paths are reachable and which state can be affected via input from untrusted domains. Such analysis would likely involve building a model of the system with a combination of taint tracking and static and symbolic analysis.

Expected results: A rigorous model of the scope of code paths and state reachable or affectable from other (Xen) domains.

Knoledge prerequisite: Frama-C, pytaint, angr, others.

Mentor: Jean-Philippe Ouellet

Audio support for Qubes Windows Tools

Project: Audio support for Qubes Windows Tools

Brief explanation: Add audio support for Windows HVMs via Qubes Windows Tools. #2624

Expected results: Windows HVMs should have an audio device that supports playback and recording.

Knowledge prerequisite: C/C++ languages, familiarity with Windows API, possibly familiarity with Windows audio stack on the driver level.

Mentor: Rafał Wojdyła

Improve Windows GUI agent performance and stability

Project: Improve Windows GUI agent performance and stability

Brief explanation: Previous profiling has shown that the Windows GUI agent uses significant portion of VM's CPU time for mouse input simulation. This can be improved, as well as agent's stability in some cases (desktop/user switching, logon/logoff, domain-joined VMs, multiple monitors). Seamless GUI experience can be significantly improved, but that may require changes in the Qubes video driver. #1044 #1045 #1500 #2138 #2487 #2589

Expected results: Reduction of agent's CPU usage, improved stability.

Knowledge prerequisite: C language, Familiarity with Windows API, especially the windowing stack. Familiarity with profiling and debugging tools for Windows.

Mentor: Rafał Wojdyła

Gui agent for Windows 8/10

Project: Gui agent for Windows 8/10

Brief explanation: Add support for Windows 8+ to the Qubes GUI agent and video driver. Starting from Windows 8, Microsoft requires all video drivers to conform to the WDDM display driver model which is incompatible with the current Qubes video driver. Unfortunately the WDDM model is much more complex than the old XPDM one and officially requires a physical GPU device (which may be emulated). Some progress has been made to create a full WDDM driver that doesn't require a GPU device, but the driver isn't working correctly yet. Alternatively, WDDM model supports display-only drivers which are much simpler but don't have access to system video memory and rendering surfaces (a key feature that would simplify seamless GUI mode). #1861

Expected results: Working display-only WDDM video driver or significant progress towards making the full WDDM driver work correctly.

Knowledge prerequisite: C/C++ languages, familiarity with Windows API, familiarity with the core Windows WDM driver model. Ideally familiarity with the WDDM display driver model.

Mentor: Rafał Wojdyła

Make Anti Evil Maid resistant against shoulder surfing and video surveillance

Project: Observing the user during early boot should not be sufficient to defeat the protection offered by Anti Evil Maid.

Brief explanation:

1. Implement optional support for time-based one-time-password seed secrets. Instead of verifying a static text or picture (which the attacker can record and replay later on a compromised system), the user would verify an ephemeral six-digit code displayed on another device, e.g. a smartphone running any Google Authenticator compatible code generator app.

2. Implement optional support for storing a passphrase-encrypted LUKS disk decryption key on a secondary AEM device. The attacker would then have to seize this device in order to decrypt the user's data; just recording the passphrase as it is entered would no longer be enough.

Expected results: AEM package updates implementing both features, with fallback support in case the user does not have their smartphone or secondary AEM device at hand. Good UX and documentation for enrolling or upgrading users.

Knowledge prerequisite:

• Bash scripting
• The AEM threat model
• GRUB2, dracut, systemd, LUKS

Mentor: Rusty Bird

GNOME support in dom0

Project: GNOME support in dom0

Brief explanation: Integrating GNOME into Qubes dom0. This include:

• patching window manager to add colorful borders
• removing stuff not needed in dom0 (file manager(s), indexing services etc)
• adjusting menu for easy navigation (same applications in different VMs and such problems, dom0-related entries in one place)
• More info: #1806

Expected results:

• Review existing support for other desktop environments (KDE, Xfce4, i3, awesome).
• Patch window manager to draw colorful borders (we use only server-side decorations), there is already very similar patch in Cappsule project.
• Configure GNOME to not make use of dom0 user home in visible way (no search in files there, no file manager, etc).
• Configure GNOME to not look into external devices plugged in (no auto mounting, device notifications etc).
• Package above modifications as rpms, preferably as extra configuration files and/or plugins than overwriting existing files. Exceptions to this rule may apply if no other option.
• Adjust comps.xml (in installer-qubes-os repo) to define package group with all required packages.
• Document installation procedure.

Knowledge prerequisite:

• GNOME architecture
• C language (patching metacity)
• Probably also javascript - for modifying GNOME shell extensions

Mentor: Marek Marczykowski-Górecki

Mitigate focus-stealing attacks

Project: Mitigate focus-stealing attacks

Brief explanation: Focus stealing attacks have long been an issue in Qubes OS. The Qubes community has long punted the issue due to having higher priority things to work on, and it being viewed as the responsability of the window manager, but nevertheless it remains a serious issue, and an effective mitigation would be most welcome. Any student wishing to work on this would need to engage the community in a discussion about the effectiveness of their proposed earlier rather than later. #1166

Expected results: Working robust focus stealing prevention for Xfce (currently the default Qubes desktop environment) or Gnome (the targeted future Qubes desktop environment).

Knoledge prerequisite: X APIs, Qubes GUI protocol, familiarity with the targeted window manager.

Mentor:

Progress towards reproducible builds

Project: Progress towards reproducible builds

Brief explanation: A long-term goal is to be able to build the entire OS and installation media in a completely bit-wise deterministic manner, but there are many baby steps to be taken along that path. See:

• "Security challenges for the Qubes build process"
• This mailing list post
• and reproducible-builds.org

for more information and qubes-specific background.

Expected results: Significant progress towards making the Qubes build process deterministic. This would likely involve cooperation with and hacking on several upstream build tools to eliminate sources of variability.

Knoledge prerequisite: qubes-builder [1] [2] [3], and efficient at introspecting complex systems: comfortable with tracing and debugging tools, ability to quickly identify and locate issues within a large codebase (upstream build tools), etc.

Mentor:

Android development in Qubes

Project: Research running Android in Qubes VM (probably HVM) and connecting it to Android Studio

Brief explanation: The goal is to enable Android development (and testing!) on Qubes OS. Currently it's only possible using qemu-emulated Android for ARM. Since it's software emulation it's rather slow. Details, reference: #2233

Expected results:

Knowledge prerequisite:

Mentor:

---

We adapted some of the language here about GSoC from the KDE GSoC page.