1441894 - Snap apparmor denial binding to dbus

Status: UNCONFIRMED

(Release Engineering :: Release Automation: Snap, defect)

Description

[Ken VanDine]

User Agent: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Ubuntu Chromium/65.0.3325.88 Chrome/65.0.3325.88 Safari/537.36

Steps to reproduce:

Running the snap generates these errors:

apparmor="DENIED" operation="dbus_bind"  bus="session" name="org.mozilla.firefox.V29yaw__" 
mask="bind" pid=25893 label="snap.firefox.firefox". It should use the dbus interface slot to bind to this path

Comment 1

[Ken VanDine]

We can use the dbus interface to allow the snap to bind to a well known dbus name.  Firefox is appending a seemingly random string to the name.  In this example it's "org.mozilla.firefox.V29yaw__".  Can anyone help answer why this needed?

Comment 2

[jlorenzo (disabled)]

I couldn't find who's may be in charge of GTK's DBus. Would you have an idea, Mike?

Comment 3

[Mike Kaply [:mkaply]]

Martin Stransky. And it appears it's this bug?

https://bugzilla.mozilla.org/show_bug.cgi?id=1418770

Comment 4

[Ken VanDine]

(In reply to Mike Kaply [:mkaply] from comment #3)
> Martin Stransky. And it appears it's this bug?
> 
> https://bugzilla.mozilla.org/show_bug.cgi?id=1418770

That shouldn't be related.  It fails to bind to the interface silently right now.  We can tell snap it's allowed to bind to the name, but I think it needs a well-known name, like org.mozilla.firefox.  I'm asking our security team if there's a way to allow a wildcard like "org.mozilla.firefox.*".

Comment 5

[Mike Kaply [:mkaply]]

I'm confused. Is this a new problem? I'm wondering why it wasn't found before.

There were multiple changes related to dbus:

https://hg.mozilla.org/mozilla-central/log/tip/widget/xremoteclient/DBusRemoteClient.cpp

This particular change (appending the profile directory) is about allowing multiple copies of Firefox to run at the same time.

Comment 6

[Ken VanDine]

I'm trying to figure out what exactly isn't working.  Currently apparmor is denying the request to bind to the dbus name.  So effectively the DBusRemoteClient stuff is just failing silently.  Everything seems to behave just fine, even when launching a second instance with a second profile.  They both seem to work fine, but I could be missing something.  I guess it's just gracefully falling back as if there is no dbus session available.  

Can you describe what functionality we lose by not having the dbus binding?

Comment 7

[Martin Stránský [:stransky] (ni? me)]

(In reply to Ken VanDine from comment #1)
> We can use the dbus interface to allow the snap to bind to a well known dbus
> name.  Firefox is appending a seemingly random string to the name.  In this
> example it's "org.mozilla.firefox.V29yaw__".  Can anyone help answer why
> this needed?

The appendix "V29yaw__" is encoded (due to DBus limitations) Firefox profile - you can have more than one active Firefox instance running.

The DBus Remote service is actively utilized by Wayland backend only - we may add an option to disable it when running under snap.

Comment 8

[Ken VanDine]

(In reply to Martin Stránský [:stransky] from comment #7)
> (In reply to Ken VanDine from comment #1)
> > We can use the dbus interface to allow the snap to bind to a well known dbus
> > name.  Firefox is appending a seemingly random string to the name.  In this
> > example it's "org.mozilla.firefox.V29yaw__".  Can anyone help answer why
> > this needed?
> 
> The appendix "V29yaw__" is encoded (due to DBus limitations) Firefox profile
> - you can have more than one active Firefox instance running.
> 
> The DBus Remote service is actively utilized by Wayland backend only - we
> may add an option to disable it when running under snap.

Our security team suggested perhaps you could use the same well-known dbus name and differentiate the profiles with different paths.  That way we could still safely confine the snap and still allowing access to bind to the dbus name.

Thoughts?

Comment 9

[Martin Stránský [:stransky] (ni? me)]

(In reply to Ken VanDine from comment #8)
> 
> Our security team suggested perhaps you could use the same well-known dbus
> name and differentiate the profiles with different paths.  That way we could
> still safely confine the snap and still allowing access to bind to the dbus
> name.
> 
> Thoughts?

Well, I can review such patch if you write one.

Comment 10

[Ken VanDine]

(In reply to Martin Stránský [:stransky] from comment #9)
> (In reply to Ken VanDine from comment #8)
> > 
> > Our security team suggested perhaps you could use the same well-known dbus
> > name and differentiate the profiles with different paths.  That way we could
> > still safely confine the snap and still allowing access to bind to the dbus
> > name.
> > 
> > Thoughts?
> 
> Well, I can review such patch if you write one.

Great, we'll work on that.

Comment 11

[Liz Henry (:lizzard Please n-i to RyanVM, jcristau, or pascal)]

Looks like this is not ready for 59.

Comment 12

[jlorenzo (disabled)]

I just confirmed with Canonical this bug is minor and has no user impact. They are okay to release 59.0 without it fixed.

Comment 13

[Olivier Tilloy]

(In reply to Ken VanDine from comment #8)

Our security team suggested perhaps you could use the same well-known dbus
name and differentiate the profiles with different paths. That way we could
still safely confine the snap and still allowing access to bind to the dbus
name.

I'm not sure how that would work: each firefox instance is tied to a given profile, and we can't have two instances claim the same well-known bus name, can we?

Comment 14

[Olivier Tilloy]

Related discussion on the snapcraft forum.

Comment 15

[Mike Kaply [:mkaply]]

Will that change make it into snapd in time for 89 or should we make a local change?

Comment 16

[Olivier Tilloy]

According to their roadmap, snapd 2.50 is scheduled for a stable release on May 12th, so it seems very unlikely that the change that I'm proposing would be accepted into the release branch before then.

What sort of local change are you thinking about?

Comment 17

[Mike Kaply [:mkaply]]

Your comment here:

"A temporary band-aid would be to have the firefox snap claim the org.mozilla.firefox.ZGVmYXVsdA__ slot, because this corresponds to the profile named “default”. This would at least make the most common use case (one single profile with the default name) work:"

Comment 18

[Olivier Tilloy]

Yes, I suppose we can do that until this is properly addressed in snapd.

Comment 19

[Olivier Tilloy]

Attachment: Bug 1441894 - allow the snap to claim the well-known D-Bus name for the "default" profile.

Comment 20

[Olivier Tilloy]

Good news, my snapd patch was merged and cherry-picked into the 2.50 branch, and I was informed that it will be part of snapd 2.50.1, scheduled for release in ~2weeks.

Comment 21

[Mike Kaply [:mkaply]]

Good news, my snapd patch was merged and cherry-picked into the 2.50 branch, and I was informed that it will be part of snapd 2.50.1, scheduled for release in ~2weeks.

Awesome! So should we wait to close this until we can test with the new snapd?

Comment 22

[Olivier Tilloy]

Awesome! So should we wait to close this until we can test with the new snapd?

Yes. Actually, this can be tested now, with snapd from the edge channel (snap refresh snapd --edge).
Just tested here, and this appears to work correctly. So I'm going to close D114774.