This is Simon McVittie's software development blog. Main site: pseudorandom.co.uk

GTK hackfest 2017: D-Bus communication with containers

At the GTK hackfest in London (which accidentally became mostly a Flatpak hackfest) I've mainly been looking into how to make D-Bus work better for app container technologies like Flatpak and Snap.

The initial motivating use cases are:

  • Portals: Portal authors need to be able to identify whether the container is being contacted by an uncontained process (running with the user's full privileges), or whether it is being contacted by a contained process (in a container created by Flatpak or Snap).

  • dconf: Currently, a contained app either has full read/write access to dconf, or no access. It should have read/write access to its own subtree of dconf configuration space, and no access to the rest.

At the moment, Flatpak runs a D-Bus proxy for each app instance that has access to D-Bus, connects to the appropriate bus on the app's behalf, and passes messages through. That proxy is in a container similar to the actual app instance, but not actually the same container; it is trusted to not pass messages through that it shouldn't pass through. The app-identification mechanism works in practice, but is Flatpak-specific, and has a known race condition due to process ID reuse and limitations in the metadata that the Linux kernel maintains for AF_UNIX sockets. In practice the use of X11 rather than Wayland in current systems is a much larger loophole in the container than this race condition, but we want to do better in future.

Meanwhile, Snap does its sandboxing with AppArmor, on kernels where it is enabled both at compile-time (Ubuntu, openSUSE, Debian, Debian derivatives like Tails) and at runtime (Ubuntu, openSUSE and Tails, but not Debian by default). Ubuntu's kernel has extra AppArmor features that haven't yet gone upstream, some of which provide reliable app identification via LSM labels, which dbus-daemon can learn by querying its AF_UNIX socket. However, other kernels like the ones in openSUSE and Debian don't have those. The access-control (AppArmor mediation) is implemented in upstream dbus-daemon, but again doesn't work portably, and is not sufficiently fine-grained or flexible to do some of the things we'll likely want to do, particularly in dconf.

After a lot of discussion with dconf maintainer Allison Lortie and Flatpak maintainer Alexander Larsson, I think I have a plan for fixing this.

This is all subject to change: see fd.o #100344 for the latest ideas.

Identity model

Each user (uid) has some uncontained processes, plus 0 or more containers.

The uncontained processes include dbus-daemon itself, desktop environment components such as gnome-session and gnome-shell, the container managers like Flatpak and Snap, and so on. They have the user's full privileges, and in particular they are allowed to do privileged things on the user's session bus (like running dbus-monitor), and act with the user's full privileges on the system bus. In generic information security jargon, they are the trusted computing base; in AppArmor jargon, they are unconfined.

The containers are Flatpak apps, or Snap apps, or other app-container technologies like Firejail and AppImage (if they adopt this mechanism, which I hope they will), or even a mixture (different app-container technologies can coexist on a single system). They are containers (or container instances) and not "apps", because in principle, you could install com.example.MyApp 1.0, run it, and while it's still running, upgrade to com.example.MyApp 2.0 and run that; you'd have two containers for the same app, perhaps with different permissions.

Each container has an container type, which is a reversed DNS name like org.flatpak or io.snapcraft representing the container technology, and an app identifier, an arbitrary non-empty string whose meaning is defined by the container technology. For Flatpak, that string would be another reversed DNS name like com.example.MyGreatApp; for Snap, as far as I can tell it would look like example-my-great-app.

The container technology can also put arbitrary metadata on the D-Bus representation of a container, again defined and namespaced by the container technology. For instance, Flatpak would use some serialization of the same fields that go in the Flatpak metadata file at the moment.

Finally, the container has an opaque container identifier identifying a particular container instance. For example, launching com.example.MyApp twice (maybe different versions or with different command-line options to flatpak run) might result in two containers with different privileges, so they need to have different container identifiers.

Contained server sockets

App-container managers like Flatpak and Snap would create an AF_UNIX socket inside the container, bind() it to an address that will be made available to the contained processes, and listen(), but not accept() any new connections. Instead, they would fd-pass the new socket to the dbus-daemon by calling a new method, and the dbus-daemon would proceed to accept() connections after the app-container manager has signalled that it has called both bind() and listen(). (See fd.o #100344 for full details.)

Processes inside the container must not be allowed to contact the AF_UNIX socket used by the wider, uncontained system - if they could, the dbus-daemon wouldn't be able to distinguish between them and uncontained processes and we'd be back where we started. Instead, they should have the new socket bind-mounted into their container's XDG_RUNTIME_DIR and connect to that, or have the new socket set as their DBUS_SESSION_BUS_ADDRESS and be prevented from connecting to the uncontained socket in some other way. Those familiar with the kdbus proposals a while ago might recognise this as being quite similar to kdbus' concept of endpoints, and I'm considering reusing that name.

Along with the socket, the container manager would pass in the container's identity and metadata, and the method would return a unique, opaque identifier for this particular container instance. The basic fields (container technology, technology-specific app ID, container ID) should probably be added to the result of GetConnectionCredentials(), and there should be a new API call to get all of those plus the arbitrary technology-specific metadata.

When a process from a container connects to the contained server socket, every message that it sends should also have the container instance ID in a new header field. This is OK even though dbus-daemon does not (in general) forbid sender-specified future header fields, because any dbus-daemon that supported this new feature would guarantee to set that header field correctly, the existing Flatpak D-Bus proxy already filters out unknown header fields, and adding this header field is only ever a reduction in privilege.

The reasoning for using the sender's container instance ID (as opposed to the sender's unique name) is for services like dconf to be able to treat multiple unique bus names as belonging to the same equivalence class of contained processes: instead of having to look up the container metadata once per unique name, dconf can look it up once per container instance the first time it sees a new identifier in a header field. For the second and subsequent unique names in the container, dconf can know that the container metadata and permissions are identical to the one it already saw.

Access control

In principle, we could have the new identification feature without adding any new access control, by keeping Flatpak's proxies. However, in the short term that would mean we'd be adding new API to set up a socket for a container without any access control, and having to keep the proxies anyway, which doesn't seem great; in the longer term, I think we'd find ourselves adding a second new API to set up a socket for a container with new access control. So we might as well bite the bullet and go for the version with access control immediately.

In principle, we could also avoid the need for new access control by ensuring that each service that will serve contained clients does its own. However, that makes it really hard to send broadcasts and not have them unintentionally leak information to contained clients - we would need to do something more like kdbus' approach to multicast, where services know who has subscribed to their multicast signals, and that is just not how dbus-daemon works at the moment. If we're going to have access control for broadcasts, it might as well also cover unicast.

The plan is that messages from containers to the outside world will be mediated by a new access control mechanism, in parallel with dbus-daemon's current support for firewall-style rules in the XML bus configuration, AppArmor mediation, and SELinux mediation. A message would only be allowed through if the XML configuration, the new container access control mechanism, and the LSM (if any) all agree it should be allowed.

By default, processes in a container can send broadcast signals, and send method calls and unicast signals to other processes in the same container. They can also receive method calls from outside the container (so that interfaces like org.freedesktop.Application can work), and send exactly one reply to each of those method calls. They cannot own bus names, communicate with other containers, or send file descriptors (which reduces the scope for denial of service).

Obviously, that's not going to be enough for a lot of contained apps, so we need a way to add more access. I'm intending this to be purely additive (start by denying everything except what is always allowed, then add new rules), not a mixture of adding and removing access like the current XML policy language.

There are two ways we've identified for rules to be added:

  • The container manager can pass a list of rules into the dbus-daemon at the time it attaches the contained server socket, and they'll be allowed. The obvious example is that an org.freedesktop.Application needs to be allowed to own its own bus name. Flatpak apps' implicit permission to talk to portals, and Flatpak metadata like org.gnome.SessionManager=talk, could also be added this way.

  • System or session services that are specifically designed to be used by untrusted clients, like the version of dconf that Allison is working on, could opt-in to having contained apps allowed to talk to them (effectively making them a generalization of Flatpak portals). The simplest such request, for something like a portal, is "allow connections from any container to contact this service"; but for dconf, we want to go a bit finer-grained, with all containers allowed to contact a single well-known rendezvous object path, and each container allowed to contact an additional object path subtree that is allocated by dconf on-demand for that app.

Initially, many contained apps would work in the first way (and in particular sockets=session-bus would add a rule that allows almost everything), while over time we'll probably want to head towards recommending more use of the second.

Related topics

Access control on the system bus

We talked about the possibility of using a very similar ruleset to control access to the system bus, as an alternative to the XML rules found in /etc/dbus-1/system.d and /usr/share/dbus-1/system.d. We didn't really come to a conclusion here.

Allison had the useful insight that the XML rules are acting like a firewall: they're something that is placed in front of potentially-broken services, and not part of the services themselves (which, as with firewalls like ufw, makes it seem rather odd when the services themselves install rules). D-Bus system services already have total control over what requests they will accept from D-Bus peers, and if they rely on the XML rules to mediate that access, they're essentially rejecting that responsibility and hoping the dbus-daemon will protect them. The D-Bus maintainers would much prefer it if system services took responsibility for their own access control (with or without using polkit), because fundamentally the system service is always going to understand its domain and its intended security model better than the dbus-daemon can.

Analogously, when a network service listens on all addresses and accepts requests from elsewhere on the LAN, we sometimes work around that by protecting it with a firewall, but the optimal resolution is to get that network service fixed to do proper authentication and access control instead.

For system services, we continue to recommend essentially this "firewall" configuration, filling in the ${} variables as appropriate:

<busconfig>
    <policy user="${the daemon uid under which the service runs}">
        <allow own="${the service's bus name}"/>
    </policy>
    <policy context="default">
        <allow send_destination="${the service's bus name}"/>
    </policy>
</busconfig>

We discussed the possibility of moving towards a model where the daemon uid to be allowed is written in the .service file, together with an opt-in to "modern D-Bus access control" that makes the "firewall" unnecessary; after some flag day when all significant system services follow that pattern, dbus-daemon would even have the option of no longer applying the "firewall" (moving to an allow-by-default model) and just refusing to activate system services that have not opted in to being safe to use without it. However, the "firewall" also protects system bus clients, and services like Avahi that are not bus-activatable, against unintended access, which is harder to solve via that approach; so this is going to take more thought.

For system services' clients that follow the "agent" pattern (BlueZ, polkit, NetworkManager, Geoclue), the correct "firewall" configuration is more complicated. At some point I'll try to write up a best-practice for these.

New header fields for the system bus

At the moment, it's harder than it needs to be to provide non-trivial access control on the system bus, because on receiving a method call, a service has to remember what was in the method call, then call GetConnectionCredentials() to find out who sent it, then only process the actual request when it has the information necessary to do access control.

Allison and I had hoped to resolve this by adding new D-Bus message header fields with the user ID, the LSM label, and other interesting facts for access control. These could be "opt-in" to avoid increasing message sizes for no reason: in particular, it is not typically useful for session services to receive the user ID, because only one user ID is allowed to connect to the session bus anyway.

Unfortunately, the dbus-daemon currently lets unknown fields through without modification. With hindsight this seems an unwise design choice, because header fields are a finite resource (there are 255 possible header fields) and are defined by the D-Bus Specification. The only field that can currently be trusted is the sender's unique name, because the dbus-daemon sets that field, overwriting the value in the original message (if any).

To make it safe to rely on the new fields, we would have to make the dbus-daemon filter out all unknown header fields, and introduce a mechanism for the service to check (during connection to the bus) whether the dbus-daemon is sufficiently new that it does so. If connected to an older dbus-daemon, the service would not be able to rely on the new fields being true, so it would have to ignore the new fields and treat them as unset. The specification is sufficiently vague that making new dbus-daemons filter out unknown header fields is a valid change (it just says that "Header fields with an unknown or unexpected field code must be ignored", without specifying who must ignore them, so having the dbus-daemon delete those fields seems spec-compliant).

This all seemed fine when we discussed it in person; but GDBus already has accessors for arbitrary header fields by numeric ID, and I'm concerned that this might mean it's too easy for a system service to be accidentally insecure: It would be natural (but wrong!) for an implementor to assume that if g_message_get_header (message, G_DBUS_MESSAGE_HEADER_FIELD_SENDER_UID) returned non-NULL, then that was guaranteed to be the correct, valid sender uid. As a result, fd.o #100317 might have to be abandoned. I think more thought is needed on that one.

Unrelated topics

As happens at any good meeting, we took the opportunity of high-bandwidth discussion to cover many useful things and several useless ones. Other discussions that I got into during the hackfest included, in no particular order:

  • .desktop file categories and how to adapt them for AppStream, perhaps involving using the .desktop vocabulary but relaxing some of the hierarchy restrictions so they behave more like "tags"
  • how to build a recommended/reference "app store" around Flatpak, aiming to host upstream-supported builds of major projects like LibreOffice
  • how Endless do their content-presenting and content-consuming apps in GTK, with a lot of "tile"-based UIs with automatic resizing and reflowing (similar to responsive design), and the applicability of similar widgets to GNOME and upstream GTK
  • whether and how to switch GNOME developer documentation to Hotdoc
  • whether pies, fish and chips or scotch eggs were the most British lunch available from Borough Market
  • the distinction between stout, mild and porter

More notes are available from the GNOME wiki.

Acknowledgements

The GTK hackfest was organised by GNOME and hosted by Red Hat and Endless. My attendance was sponsored by Collabora. Thanks to all the sponsors and organisers, and the developers and organisations who attended.

GTK Hackfest 2016

I'm back from the GTK hackfest in Toronto, Canada and mostly recovered from jetlag, so it's time to write up my notes on what we discussed there.

Despite the hackfest's title, I was mainly there to talk about non-GUI parts of the stack, and technologies that fit more closely in what could be seen as the freedesktop.org platform than they do in GNOME. In particular, I'm interested in Flatpak as a way to deploy self-contained "apps" in a freedesktop-based, sandboxed runtime environment layered over the Universal Operating System and its many derivatives, with both binary and source compatibility with other GNU/Linux distributions.

I'm mainly only writing about discussions I was directly involved in: lots of what sounded like good discussion about the actual graphics toolkit went over my head completely :-) More notes, mostly from Matthias Clasen, are available on the GNOME wiki.

In no particular order:

Thinking with portals

We spent some time discussing Flatpak's portals, mostly on Tuesday. These are the components that expose a subset of desktop functionality as D-Bus services that can be used by contained applications: they are part of the security boundary between a contained app and the rest of the desktop session. Android's intents are a similar concept seen elsewhere. While the portals are primarily designed for Flatpak, there's no real reason why they couldn't be used by other app-containment solutions such as Canonical's Snap.

One major topic of discussion was their overall design and layout. Most portals will consist of a UX-independent part in Flatpak itself, together with a UX-specific implementation of any user interaction the portal needs. For example, the portal for file selection has a D-Bus service in Flatpak, which interacts with some UX-specific service that will pop up a standard UX-specific "Open" dialog — for GNOME and probably other GTK environments, that dialog is in (a branch of) GTK.

A design principle that was reiterated in this discussion is that the UX-independent part should do as much as possible, with the UX-specific part only carrying out the user interactions that need to comply with a particular UX design (in the GTK case, GNOME's design). This minimizes the amount of work that needs to be redone for other desktop or embedded environments, while still ensuring that the other environments can have their chosen UX design. In particular, it's important that, as much as possible, the security- and performance-sensitive work (such as data transport and authentication) is shared between all environments.

The aim is for portals to get the user's permission to carry out actions, while keeping it as implicit as possible, avoiding an "are you sure?" step where feasible. For example, if an application asks to open a file, the user's permission is implicitly given by them selecting the file in the file-chooser dialog and pressing OK: if they do not want this application to open a file at all, they can deny permission by cancelling. Similarly, if an application asks to stream webcam data, the UX we expect is for GNOME's Cheese app (or a similar non-GNOME app) to appear, open the webcam to provide a preview window so they can see what they are about to send, but not actually start sending the stream to the requesting app until the user has pressed a "Start" button. When defining the API "contracts" to be provided by applications in that situation, we will need to be clear about whether the provider is expected to obtain confirmation like this: in most cases I would anticipate that it is.

One security trade-off here is that we have to have a small amount of trust in the providing app. For example, continuing the example of Cheese as a webcam provider, Cheese could (and perhaps should) be a contained app itself, whether via something like Flatpak, an LSM like AppArmor or both. If Cheese is compromised somehow, then whenever it is running, it would be technically possible for it to open the webcam, stream video and send it to a hostile third-party application. We concluded that this is an acceptable trade-off: each application needs to be trusted with the privileges that it needs to do its job, and we should not put up barriers that are easy to circumvent or otherwise serve no purpose.

The main (only?) portal so far is the file chooser, in which the contained application asks the wider system to show an "Open..." dialog, and if the user selects a file, it is returned to the contained application through a FUSE filesystem, the document portal. The reference implementation of the UX for this is in GTK, and is basically a GtkFileChooserDialog. The intention is that other environments such as KDE will substitute their own equivalent.

Other planned portals include:

  • image capture (scanner/camera)
  • opening a specified URI
    • this needs design feedback on how it should work for non-http(s)
  • sharing content, for example on social networks (like Android's Sharing menu)
  • proxying joystick/gamepad input (perhaps via Wayland or FUSE, or perhaps by modifying libraries like SDL with a new input source)
  • network proxies (GProxyResolver) and availability (GNetworkMonitor)
  • contacts/address book, probably vCard-based
  • notifications, probably based on freedesktop.org Notifications
  • video streaming (perhaps using Pinot, analogous to PulseAudio but for video)

Environment variables

GNOME on Wayland currently has a problem with environment variables: there are some traditional ways to set environment variables for X11 sessions or login shells using shell script fragments (/etc/X11/Xsession.d, /etc/X11/xinit/xinitrc.d, /etc/profile.d), but these do not apply to Wayland, or to noninteractive login environments like cron and systemd --user. We are also keen to avoid requiring a Turing-complete shell language during session startup, because it's difficult to reason about and potentially rather inefficient.

Some uses of environment variables can be dismissed as unnecessary or even unwanted, similar to the statement in Debian Policy §9.9: "A program must not depend on environment variables to get reasonable defaults." However, there are two common situations where environment variables can be necessary for proper OS integration: search-paths like $PATH, $XDG_DATA_DIRS and $PYTHONPATH (particularly necessary for things like Flatpak), and optionally-loaded modules like $GTK_MODULES and $QT_ACCESSIBILITY where a package influences the configuration of another package.

There is a stopgap solution in GNOME's gdm display manager, /usr/share/gdm/env.d, but this is gdm-specific and insufficiently expressive to provide the functionality needed by Flatpak: "set XDG_DATA_DIRS to its specified default value if unset, then add a couple of extra paths".

pam_env comes closer — PAM is run at every transition from "no user logged in" to "user can execute arbitrary code as themselves" — but it doesn't support .d fragments, which are required if we want distribution packages to be able to extend search paths. pam_env also turns off per-user configuration by default, citing security concerns.

I'll write more about this when I have a concrete proposal for how to solve it. I think the best solution is probably a PAM module similar to pam_env but supporting .d directories, either by modifying pam_env directly or out-of-tree, combined with clarifying what the security concerns for per-user configuration are and how they can be avoided.

Relocatable binary packages

On Windows and OS X, various GLib APIs automatically discover where the application binary is located and use search paths relative to that; for example, if C:\myprefix\bin\app.exe is running, GLib might put C:\myprefix\share into the result of g_get_system_data_dirs(), so that the application can ask to load app/data.xml from the data directories and get C:\myprefix\share\app\data.xml. We would like to be able to do the same on Linux, for example so that the apps in a Flatpak or Snap package can be constructed from RPM or dpkg packages without needing to be recompiled for a different --prefix, and so that other third-party software packages like the games on Steam and gog.com can easily locate their own resources.

Relatedly, there are currently no well-defined semantics for what happens when a .desktop file or a D-Bus .service file has Exec=./bin/foo. The meaning of Exec=foo is well-defined (it searches $PATH) and the meaning of Exec=/opt/whatever/bin/foo is obvious. When this came up in D-Bus previously, my assertion was that the meaning should be the same as in .desktop files, whatever that is.

We agreed to propose that the meaning of a non-absolute path in a .desktop or .service file should be interpreted relative to the directory where the .desktop or .service file was found: for example, if /opt/whatever/share/applications/foo.desktop says Exec=../../bin/foo, then /opt/whatever/bin/foo would be the right thing to execute. While preparing a mail to the freedesktop and D-Bus mailing lists proposing this, I found that I had proposed the same thing almost 2 years ago... this time I hope I can actually make it happen!

Flatpak and OSTree bug fixing

On the way to the hackfest, and while the discussion moved to topics that I didn't have useful input on, I spent some time fixing up the Debian packaging for Flatpak and its dependencies. In particular, I did my first upload as a co-maintainer of bubblewrap, uploaded ostree to unstable (with the known limitation that the grub, dracut and systemd integration is missing for now since I haven't been able to test it yet), got most of the way through packaging Flatpak 0.6.5 (which I'll upload soon), cherry-picked the right patches to make ostree compile on Debian 8 in an effort to make backports trivial, and spent some time disentangling a flatpak test failure which was breaking the Debian package's installed-tests. I'm still looking into ostree test failures on little-endian MIPS, which I was able to reproduce on a Debian porterbox just before the end of the hackfest.

OSTree + Debian

I also had some useful conversations with developers from Endless, who recently opened up a version of their OSTree build scripts for public access. Hopefully that information brings me a bit closer to being able to publish a walkthrough for how to deploy a simple Debian derivative using OSTree (help with that is very welcome of course!).

GTK life-cycle and versioning

The life-cycle of GTK releases has already been mentioned here and elsewhere, and there are some interesting responses in the comments on my earlier blog post.

It's important to note that what we discussed at the hackfest is only a proposal: a hackfest discussion between a subset of the GTK maintainers and a small number of other GTK users (I am in the latter category) doesn't, and shouldn't, set policy for all of GTK or for all of GNOME. I believe the intention is that the GTK maintainers will discuss the proposals further at GUADEC, and make a decision after that.

As I said before, I hope that being more realistic about API and ABI guarantees can avoid GTK going too far towards either of the possible extremes: either becoming unable to advance because it's too constrained by compatibility, or breaking applications because it isn't constrained enough. The current situation, where it is meant to be compatible within the GTK 3 branch but in practice applications still sometimes break, doesn't seem ideal for anyone, and I hope we can do better in future.

Acknowledgements

Thanks to everyone involved, particularly:

  • Matthias Clasen, who organised the hackfest and took a lot of notes
  • Allison Lortie, who provided on-site cat-herding and led us to some excellent restaurants
  • Red Hat Inc., who provided the venue (a conference room in their Toronto office), snacks, a lot of coffee, and several participants
  • my employers Collabora Ltd., who sponsored my travel and accomodation
GTK versioning and distributions

Allison Lortie has provoked a lot of comment with her blog post on a new proposal for how GTK is versioned. Here's some more context from the discussion at the GTK hackfest that prompted that proposal: there's actually quite a close analogy in how new Debian versions are developed.

The problem we're trying to address here is the two sides of a trade-off:

  • Without new development, a library (or an OS) can't go anywhere new
  • New development sometimes breaks existing applications

Historically, GTK has aimed to keep compatible within a major version, where major versions are rather far apart (GTK 1 in 1998, GTK 2 in 2002, GTK 3 in 2011, GTK 4 somewhere in the future). Meanwhile, fixing bugs, improving performance and introducing new features sometimes results in major changes behind the scenes. In an ideal world, these behind-the-scenes changes would never break applications; however, the world isn't ideal. (The Debian analogy here is that as much as we aspire to having the upgrade from one stable release to the next not break anything at all, I don't think we've ever achieved that in practice - we still ask users to read the release notes, even though ideally that wouldn't be necessary.)

In particular, the perceived cost of doing a proper ABI break (a fully parallel-installable GTK 4) means there's a strong temptation to make changes that don't actually remove or change C symbols, but are clearly an ABI break, in the sense that an application that previously worked and was considered correct no longer works. A prominent recent example is the theming changes in GTK 3.20: the ABI in terms of functions available didn't change, but what happens when you call those functions changed in an incompatible way. This makes GTK hard to rely on for applications outside the GNOME release cycle, which is a problem that needs to be fixed (without stopping development from continuing).

The goal of the plan we discussed today is to decouple the latest branch of development, which moves fast and sometimes breaks API, from the API-stable branches, which only get bug fixes. This model should look quite familiar to Debian contributors, because it's a lot like the way we release Debian and Ubuntu.

In Debian, at any given time we have a development branch (testing/unstable) - currently "stretch", the future Debian 9. We also have some stable branches, of which the most recent are Debian 8 "jessie" and Debian 7 "wheezy". Different users of Debian have different trade-offs that lead them to choose one or the other of these. Users who value stability and want to avoid unexpected changes, even at a cost in terms of features and fixes for non-critical bugs, choose to use a stable release, preferably the most recent; they only need to change what they run on top of Debian for OS API changes (for instance webapps, local scripts, or the way they interact with the GUI) approximately every 2 years, or perhaps less often than that with the Debian-LTS project supporting non-current stable releases. Meanwhile, users who value the latest versions and are willing to work with a "moving target" as a result choose to use testing/unstable.

The GTK analogy here is really quite close. In the new versioning model, library users who value stability over new things would prefer to use a stable-branch, ideally the latest; library users who want the latest features, the latest bug-fixes and the latest new bugs would use the branch that's the current focus of development. In practice we expect that the latter would be mostly GNOME projects. There's been some discussion at the hackfest about how often we'd have a new stable-branch: the fastest rate that's been considered is a stable-branch every 2 years, similar to Ubuntu LTS and Debian, but there's no consensus yet on whether they will be that frequent in practice.

How many stable versions of GTK would end up shipped in Debian depends on how rapidly projects move from "old-stable" to "new-stable" upstream, how much those projects' Debian maintainers are willing to patch them to move between branches, and how many versions the release team will tolerate. Once we reach a steady state, I'd hope that we might have 1 or 2 stable-branched versions active at a time, packaged as separate parallel-installable source packages (a lot like how we handle Qt). GTK 2 might well stay around as an additional active version just from historical inertia. The stable versions are intended to be fully parallel-installable, just like the situation with GTK 1.2, GTK 2 and GTK 3 or with the major versions of Qt.

For the "current development" version, I'd anticipate that we'd probably only ship one source package, and do ABI transitions for one version active at a time, a lot like how we deal with libgnome-desktop and the evolution-data-server family of libraries. Those versions would have parallel-installable runtime libraries but non-parallel-installable development files, again similar to libgnome-desktop.

At the risk of stretching the Debian/Ubuntu analogy too far, the intermediate "current development" GTK releases that would accompany a GNOME release are like Ubuntu's non-LTS suites: they're more up to date than the fully stable releases (Ubuntu LTS, which has a release schedule similar to Debian stable), but less stable and not supported for as long.

Hopefully this plan can meet both of its goals: minimize breakage for applications, while not holding back the development of new APIs.

Flatpak in Debian

Quite a lot has happened in xdg-app since last time I blogged about it. Most noticeably, it isn't called xdg-app any more, having been renamed to Flatpak. It is now available in Debian experimental under that name, and the xdg-app package that was briefly there has been removed. I'm currently in the process of updating Flatpak to the latest version 0.6.4.

The privileged part has also spun off into a separate project, Bubblewrap, which recently had its first release (0.1.0). This is intended as a common component with which unprivileged users can start a container in a way that won't let them escalate privileges, like a more flexible version of linux-user-chroot.

Bubblewrap has also been made available in Debian, maintained by Laszlo Boszormenyi (also maintainer of linux-user-chroot). Yesterday I sent a patch to update Laszlo's packaging for 0.1.0. I'm hoping to become a co-maintainer to upload that myself, since I suspect Flatpak and Bubblewrap might need to track each other quite closely. For the moment, Flatpak still uses its own internal copy of Bubblewrap, but I consider that to be a bug and I'd like to be able to fix it soon.

At some point I also want to experiment with using Bubblewrap to sandbox some of the game engines that are packaged in Debian: networked games are a large attack surface, and typically consist of the sort of speed-optimized C or C++ code that is an ideal home for security vulnerabilities. I've already made some progress on jailing game engines with AppArmor, but making sensitive files completely invisible to the game engine seems even better than preventing them from being opened.

Next weekend I'm going to be heading to Toronto for the GTK Hackfest, primarily to to talk to GNOME and Flatpak developers about their plans for sandboxing, portals and Flatpak. Hopefully we can make some good progress there: the more I know about the state of software security, the less happy I am with random applications all being equally privileged. Newer display technologies like Wayland and Mir represent an opportunity to plug one of the largest holes in typical application containerization, which is a major step in bringing sandboxes like Flatpak and Snap from proof-of-concept to a practical improvement in security.

Other next steps for Flatpak in Debian:

  • To get into the next stable release (Debian 9), Flatpak needs to move from experimental into unstable and testing. I've taken the first step towards that by uploading libgsystem to unstable. Before Flatpak can follow, OSTree also needs to move.
  • Now that it's in Debian, please report bugs in the usual Debian way or send patches to fix bugs: Flatpak, OSTree, libgsystem.
  • In particular, there are some OSTree bugs tagged help. I'd appreciate contributions to the OSTree packaging from people who are interested in using it to deploy dpkg-based operating systems - I'm primarily looking at it from the Flatpak perspective, so the boot/OS side of it isn't so well tested. Red Hat have rpm-ostree, and I believe Endless do something analogous to build OS images with dpkg, but I haven't had a chance to look into that in detail yet.
  • Co-maintainers for Flatpak, OSTree, libgsystem would also be very welcome.
GNOME Developer Experience hackfest: xdg-app + Debian

Over the last few days I've been at the GNOME Developer Experience hackfest in Brussels, looking into xdg-app and how best to use it in Debian and Debian derivatives.

xdg-app is basically a way to run "non-core" software on Linux distributions, analogous to apps on Android and iOS. It doesn't replace distributions like Debian or packaging systems, but it adds a layer above them. It's mostly aimed towards third-party apps obtained from somewhere that isn't your distribution vendor, aiming to address a few long-standing problems in that space:

  • There's no single ABI that can be called "a standard Linux system" in the same way there would be for Windows or OS X or Android or whatever, apart from LSB which is rather limited. Testing that a third-party app "works on Linux", or even "works on stable Linux releases from 2015", involves a combinatorial explosion of different distributions, desktop environments and local configurations. Steam uses the Steam Runtime, a chroot environment closely resembling Ubuntu 12.04 LTS; other vendors tend to test on a vaguely recent Ubuntu LTS and leave it at that.

  • There's no widely-supported mechanism for installing third-party applications as an ordinary user. gog.com used to distribute Ubuntu- and Debian-compatible .deb files, but installing a .deb involves running arbitrary vendor-supplied scripts as root, which should worry anyone who wants any sort of privilege-separation. (They have now switched to executable self-extracting installers, which involve running arbitrary vendor-supplied scripts as an ordinary user... better, but not perfect.)

  • Relatedly, the third-party application itself runs with the user's full privileges: a malicious or security-buggy third-party application can do more or less anything, unless you either switch to a different uid to run third-party apps, or use a carefully-written, app-specific AppArmor profile or equivalent.

To address the first point, each application uses a specified "runtime", which is available as /usr inside its sandbox. This can be used to run application bundles with multiple, potentially incompatible sets of dependencies within the same desktop environment. A runtime can be updated within its branch - for instance, if an application uses the "GNOME 3.18" runtime (consisting of a basic Linux system, the GNOME 3.18 libraries, other related libraries like Mesa, and their recursive dependencies like libjpeg), it can expect to see minor-version updates from GNOME 3.18.x (including any security updates that might be necessary for the bundled libraries), but not a jump to GNOME 3.20.

To address the second issue, the plan is for application bundles to be available as a single file, containing metadata (such as the runtime to use), the app itself, and any dependencies that are not available in the runtime (which the app vendor is responsible for updating if necessary). However, the primary way to distribute and upgrade runtimes and applications is to package them as OSTree repositories, which provide a git-like content-addressed filesystem, with efficient updates using binary deltas. The resulting files are hard-linked into place.

To address the last point, application bundles run partially isolated from the wider system, using containerization techniques such as namespaces to prevent direct access to system resources. Resources from outside the sandbox can be accessed via "portal" services, which are responsible for access control; for example, the Documents portal (the only one, so far) displays an "Open" dialog outside the sandbox, then allows the application to access only the selected file.

xdg-app for Debian

One thing I've been doing at this hackfest is improving the existing Debian/Ubuntu packaging for xdg-app (and its dependencies ostree and libgsystem), aiming to get it into a state where I can upload it to Debian experimental. Because xdg-app aims to be a general freedesktop project, I'm currently intending to make it part of the "Utopia" packaging team alongside projects like D-Bus and polkit, but I'm open to suggestions if people want to co-maintain it elsewhere.

In the process of updating xdg-app, I sent various patches to Alex, mostly fixing build and test issues, which are in the new 0.4.8 release.

I'd appreciate co-maintainers and further testing for this stuff, particularly ostree: ostree is primarily a whole-OS deployment technology, which isn't a use-case that I've tested, and in particular ostree-grub2 probably doesn't work yet.

Source code:

Binaries (no trust path, so only use these if you have a test VM):

  • deb https://people.debian.org/~smcv/xdg-app xdg-app main

The "Hello, World" of xdg-apps

Another thing I set out to do here was to make a runtime and an app out of Debian packages. Most of the test applications in and around GNOME use the "freedesktop" or "GNOME" runtimes, which consist of a Yocto base system and lots of RPMs, are rebuilt from first principles on-demand, and are extensive and capable enough that they make it somewhat non-obvious what's in an app or a runtime.

So, here's a step-by-step route through xdg-app, first using typical GNOME instructions, but then using the simplest GUI app I could find - xvt, a small xterm clone. I'm using a Debian testing (stretch) x86_64 virtual machine for all this. xdg-app currently requires systemd-logind to put users and apps in cgroups, either with systemd as pid 1 (systemd-sysv) or systemd-shim and cgmanager; I used the default systemd-sysv. In principle it could work with plain cgmanager, but nobody has contributed that support yet.

Demonstrating an existing xdg-app

Debian's kernel is currently patched to be able to allow unprivileged users to create user namespaces, but make it runtime-configurable, because there have been various security issues in that feature, making it a security risk for a typical machine (and particularly a server). Hopefully unprivileged user namespaces will soon be secure enough that we can enable them by default, but for now, we have to do one of three things to let xdg-app use them:

  • enable unprivileged user namespaces via sysctl:

    sudo sysctl kernel.unprivileged_userns_clone=1
    
  • make xdg-app root-privileged (it will keep CAP_SYS_ADMIN and drop the rest):

    sudo dpkg-statoverride --update --add root root 04755 /usr/bin/xdg-app-helper
    
  • make xdg-app slightly less privileged:

    sudo setcap cap_sys_admin+ep /usr/bin/xdg-app-helper
    

First, we'll need a runtime. The standard xdg-app tutorial would tell you to download the "GNOME Platform" version 3.18. To do that, you'd add a remote, which is a bit like a git remote, and a bit like an apt repository:

$ wget http://sdk.gnome.org/keys/gnome-sdk.gpg
$ xdg-app remote-add --user --gpg-import=gnome-sdk.gpg gnome \
    http://sdk.gnome.org/repo/

(I'm ignoring considerations like trust paths and security here, for brevity; in real life, you'd want to obtain the signing key via https and/or have a trust path to it, just like you would for a secure-apt signing key.)

You can list what's available in a remote:

$ xdg-app remote-ls --user gnome
...
org.freedesktop.Platform
...
org.freedesktop.Platform.Locale.cy
...
org.freedesktop.Sdk
...
org.gnome.Platform
...

The Platform runtimes are what we want here: they are collections of runtime libraries with which you can run an application. The Sdk runtimes add development tools, header files, etc. to be able to compile apps that will be compatible with the Platform.

For now, all we want is the GNOME 3.18 platform:

$ xdg-app install --user gnome org.gnome.Platform 3.18

Next, we can install an app that uses it, from Alex Larsson's nightly builds of a subset of GNOME. The server they're on doesn't have a great deal of bandwidth, so be nice :-)

$ wget http://209.132.179.2/keys/nightly.gpg
$ xdg-app remote-add --user --gpg-import=nightly.gpg nightly \
    http://209.132.179.2/repo/
$ xdg-app install --user nightly org.mypaint.MypaintDevel

We now have one app, and the runtime it needs:

$ xdg-app list
org.mypaint.MypaintDevel
$ xdg-app run org.mypaint.MypaintDevel
[you see a GUI window]

Digression: what's in a runtime?

Behind the scenes, xdg-app runtimes and apps are both OSTree trees. This means the ostree tool, from the package of the same name, can be used to inspect them.

$ sudo apt install ostree
$ ostree refs --repo ~/.local/share/xdg-app/repo
gnome:runtime/org.gnome.Platform/x86_64/3.18
nightly:app/org.mypaint.MypaintDevel/x86_64/master

A "ref" has roughly the same meaning as in git: something like a branch or a tag. ostree can list the directory tree that it represents:

$ ostree ls --repo ~/.local/share/xdg-app/repo \
    runtime/org.gnome.Platform/x86_64/3.18
d00755 0 0      0 /
-00644 0 0    493 /metadata
d00755 0 0      0 /files
$ ostree ls --repo ~/.local/share/xdg-app/repo \
    runtime/org.gnome.Platform/x86_64/3.18 /files
d00755 0 0      0 /files
l00777 0 0      0 /files/local -> ../var/usrlocal
l00777 0 0      0 /files/sbin -> bin
d00755 0 0      0 /files/bin
d00755 0 0      0 /files/cache
d00755 0 0      0 /files/etc
d00755 0 0      0 /files/games
d00755 0 0      0 /files/include
d00755 0 0      0 /files/lib
d00755 0 0      0 /files/lib64
d00755 0 0      0 /files/libexec
d00755 0 0      0 /files/share
d00755 0 0      0 /files/src

You can see that /files in a runtime is basically a copy of /usr. This is not coincidental: the runtime's /files gets mounted at /usr inside the xdg-app container. There is also some metadata, which is in the ini-like syntax seen in .desktop files:

$ ostree cat --repo ~/.local/share/xdg-app/repo \
    runtime/org.gnome.Platform/x86_64/3.18 /metadata
[Runtime]
name=org.gnome.Platform/x86_64/3.16
runtime=org.gnome.Platform/x86_64/3.16
sdk=org.gnome.Sdk/x86_64/3.16

[Extension org.freedesktop.Platform.GL]
version=1.2
directory=lib/GL

[Extension org.freedesktop.Platform.Timezones]
version=1.2
directory=share/zoneinfo

[Extension org.gnome.Platform.Locale]
directory=share/runtime/locale
subdirectories=true

[Environment]
GI_TYPELIB_PATH=/app/lib/girepository-1.0
GST_PLUGIN_PATH=/app/lib/gstreamer-1.0
LD_LIBRARY_PATH=/app/lib:/usr/lib/GL

Looking at an app, the situation is fairly similar:

$ ostree ls --repo ~/.local/share/xdg-app/repo \
    app/org.mypaint.MypaintDevel/x86_64/master
d00755 0 0      0 /
-00644 0 0    258 /metadata
d00755 0 0      0 /export
d00755 0 0      0 /files

This time, /files maps to what will become /app for the application, which was compiled with --prefix=/app:

$ ostree ls --repo ~/.local/share/xdg-app/repo \
    app/org.mypaint.MypaintDevel/x86_64/master /files
d00755 0 0      0 /files
-00644 0 0   4599 /files/manifest.json
d00755 0 0      0 /files/bin
d00755 0 0      0 /files/lib
d00755 0 0      0 /files/share

There is also a /export directory, which is made visible to the host system so that the contained app can appear as a "first-class citizen" in menus:

$ ostree ls --repo ~/.local/share/xdg-app/repo \
    app/org.mypaint.MypaintDevel/x86_64/master /export
d00755 0 0      0 /export
d00755 0 0      0 /export/share
user@debian:~$ ostree ls --repo ~/.local/share/xdg-app/repo \
    app/org.mypaint.MypaintDevel/x86_64/master /export/share
d00755 0 0      0 /export/share
d00755 0 0      0 /export/share/app-info
d00755 0 0      0 /export/share/applications
d00755 0 0      0 /export/share/icons
user@debian:~$ ostree ls --repo ~/.local/share/xdg-app/repo \
    app/org.mypaint.MypaintDevel/x86_64/master /export/share/applications
d00755 0 0      0 /export/share/applications
-00644 0 0    715 /export/share/applications/org.mypaint.MypaintDevel.desktop
$ ostree cat --repo ~/.local/share/xdg-app/repo \
    app/org.mypaint.MypaintDevel/x86_64/master \
    /export/share/applications/org.mypaint.MypaintDevel.desktop
[Desktop Entry]
Version=1.0
Name=(Nightly) MyPaint
TryExec=mypaint
Exec=mypaint %f
Comment=Painting program for digital artists
...
Comment[zh_HK]=藝術家的电脑绘画
GenericName=Raster Graphics Editor
GenericName[fr]=Éditeur d'Image Matricielle
MimeType=image/openraster;image/png;image/jpeg;
Type=Application
Icon=org.mypaint.MypaintDevel
StartupNotify=true
Categories=Graphics;GTK;2DGraphics;RasterGraphics;
Terminal=false

Again, there's some metadata:

$ ostree cat --repo ~/.local/share/xdg-app/repo \
    app/org.mypaint.MypaintDevel/x86_64/master /metadata
[Application]
name=org.mypaint.MypaintDevel
runtime=org.gnome.Platform/x86_64/3.18
sdk=org.gnome.Sdk/x86_64/3.18
command=mypaint

[Context]
shared=ipc;
sockets=x11;pulseaudio;
filesystems=host;

[Extension org.mypaint.MypaintDevel.Debug]
directory=lib/debug

Building a runtime, probably the wrong way

The way in which the reference/demo runtimes and containers are generated is... involved. As far as I can tell, there's a base OS built using Yocto, and the actual GNOME bits come from RPMs. However, we don't need to go that far to get a working runtime.

In preparing this runtime I'm probably completely ignoring some best-practices and tools - but it works, so it's good enough.

First we'll need a repository:

$ sudo install -d -o$(id -nu) /srv/xdg-apps
$ ostree init --repo /srv/xdg-apps

I'm just keeping this local for this demonstration, but you could rsync it to a web server's exported directory or something - a lot like a git repository, it's just a collection of files. We want everything in /usr because that's what xdg-app expects, hence usrmerge:

$ sudo mount -t tmpfs -o mode=0755 tmpfs /mnt
$ sudo debootstrap --arch=amd64 --include=libx11-6,usrmerge \
    --variant=minbase stretch /mnt http://192.168.122.1:3142/debian
$ sudo mkdir /mnt/runtime
$ sudo mv /mnt/usr /mnt/runtime/files

This obviously has a lot of stuff in it that we don't need - most obviously init, apt and dpkg - but it's Good Enough™.

We will also need some metadata. This is sufficient:

$ sudo sh -c 'cat > /mnt/runtime/metadata'
[Runtime]
name=org.debian.Debootstrap/x86_64/8.20160130
runtime=org.debian.Debootstrap/x86_64/8.20160130

That's a runtime. We can commit it to ostree, and generate xdg-app metadata:

$ ostree commit --repo /srv/xdg-apps \
    --branch runtime/org.debian.Debootstrap/x86_64/8.20160130 \
    /mnt/runtime
$ fakeroot ostree commit --repo /srv/xdg-apps \
    --branch runtime/org.debian.Debootstrap/x86_64/8.20160130
$ fakeroot xdg-app build-update-repo /srv/xdg-apps

(I'm not sure why ostree and xdg-app report "Operation not permitted" when we aren't root or fakeroot - feedback welcome.)

build-update-repo would presumably also be the right place to GPG-sign your repository, if you were doing that.

We can add that as another xdg-app remote:

$ xdg-app remote-add --user --no-gpg-verify local file:///srv/xdg-apps
$ xdg-app remote-ls --user local
org.debian.Debootstrap

Building an app, probably the wrong way

The right way to build an app is to build a "SDK" runtime - similar to that platform runtime, but with development files and tools - and recompile the app and any missing libraries with ./configure --prefix=/app && make && make install. I'm not going to do that, because simplicity is nice, and I'm reasonably sure xvt doesn't actually hard-code /usr into the binary:

$ install -d xvt-app/files/bin
$ sudo apt-get --download-only install xvt
$ dpkg-deb --fsys-tarfile /var/cache/apt/archives/xvt_2.1-20.1_amd64.deb \
    | tar -xvf - ./usr/bin/xvt
./usr/
./usr/bin/
./usr/bin/xvt
...
$ mv usr xvt-app/files

Again, we'll need metadata, and it's much simpler than the more production-quality GNOME nightly builds:

$ cat > xvt-app/metadata
[Application]
name=org.debian.packages.xvt
runtime=org.debian.Debootstrap/x86_64/8.20160130
command=xvt

[Context]
sockets=x11;
$ fakeroot ostree commit --repo /srv/xdg-apps \
    --branch app/org.debian.packages.xvt/x86_64/2.1-20.1 xvt-app
$ fakeroot xdg-app build-update-repo /srv/xdg-apps
Updating appstream branch
No appstream data for runtime/org.debian.Debootstrap/x86_64/8.20160130
No appstream data for app/org.debian.packages.xvt/x86_64/2.1-20.1
Updating summary
$ xdg-app remote-ls --user local
org.debian.Debootstrap
org.debian.packages.xvt

The obligatory screenshot

OK, good, now we can install it:

$ xdg-app install --user local org.debian.Debootstrap 8.20160130
$ xdg-app install --user local org.debian.packages.xvt 2.1-20.1
$ xdg-app run --branch=2.1-20.1 org.debian.packages.xvt

xvt in a container

and you can play around with the shell in the xvt and see what you can and can't do in the container.

I'm sure there were better ways to do most of this, but I think there's value in having such a simplistic demo to go alongside the various GNOMEish apps.


Acknowledgements:

Thanks to all those!

Older posts:

Discworld Noir in a Windows 98 virtual machine on Linux
Posted
Why polkit (or, how to mount a disk on modern Linux)
Posted
still aiming to be the universal operating system
Posted
Debian activities
Posted
Announcing gfcombinefs
Posted
RC Bugs of W48-W49
Posted
RC Bugs of W47
Posted
telepathy-qt4 0.2.0: first shared library release
Posted
DSpam and how not to do it
Posted
Converting Debian packaging from bzr to git
Posted
Why you shouldn't block on D-Bus calls
Posted
Encrypted root filesystem on a Debian laptop
Posted
Compiling against uninstalled versions of libraries
Posted
Integrating Enemies of Carlotta with Postfix
Posted
Speeding up builds with ccache and icecc
Posted
Spec-writing On A Plane
Posted
\o/
Posted
Implementing D-Bus services with telepathy-glib
Posted
Here's how it works
Posted
A magical xrandr incantation
Posted

This blog is powered by ikiwiki.