804592 - CreatePipe from Chromium IPC requires Logon Session, fails otherwise

Status: RESOLVED FIXED

(Core :: IPC, defect)

Description

[rmkn85]

When running XulRunner based application (using GeckoFX https://bitbucket.org/geckofx/geckofx-14.0) from a Windows Service, on Windows Server 2008 R2 Datacenter (meaning it's without a Local Logon Session), it crashes.

The last error from Mozilla is:
WARNING: file e:/builds/moz2_slave/rel-m-rel-xr-w32-bld/build/ipc/chromium/src/chrome/common/ipc_channel_win.cc, line 153

According to Mozilla IPC code imported from Chromium (2006-2008), it requires a Security Descriptor, and gets that through GetLogonSessionOnlyDACL(), which fails when there's no Logon Session:
http://lxr.mozilla.org/mozilla-central/source/ipc/chromium/src/chrome/common/ipc_channel_win.cc#142

The latest Chromium IPC code (2012) is different and doesn't require a Security Descriptor to create a pipe:
http://src.chromium.org/svn/trunk/src/ipc/ipc_channel_win.cc

Can latest Chromium IPC code be back-ported to Mozilla repository?

Comment 1

[Josh Matthews [:jdm]]

My suspicion is that if this isn't biting us in released versions of Firefox, you're going to need to do the work required here. The Chromium code is updated one absolutely-required bugfix at a time, these days.

Comment 2

[rmkn85]

Possibly related to these Mozilla Crash Reports
https://crash-stats.mozilla.com/report/list?signature=win_util%3A%3AGetLogonSessionOnlyDACL(_SECURITY_DESCRIPTOR**)

@Josh why isn't the Chromium code updated more regularly, including backporting fixes done in their repository?

Comment 3

[Josh Matthews [:jdm]]

Effort vs. reward. At this point we are multiple years behind, and what we have seems to be working well. It was never intended to be a permanent dependency, but at some point we gave up pretending that it was anything else.

Comment 4

[rmkn85]

The main difference relevant to this bug is the last parameter of CreateNamedPipeW().
In the code in Mozilla, it's &security_attributes, which uses GetLogonSessionOnlyDACL, which may fail (when there's no Logon Session).
In the code in Chromium, they are just passing NULL to that parameter.

According to Microsoft Documentation, passing NULL gets default security descriptor.
http://msdn.microsoft.com/en-us/library/windows/desktop/aa365150(v=vs.85).aspx

Is there any reason not to do this modification?

Comment 5

[Josh Matthews [:jdm]]

We've approved similar patches before. Would you mind creating one?

Comment 6

[rmkn85]

I haven't submitted any patches to Mozilla before, does it mean I need to download the entire source code, build it and run all the tests, to submit this patch for approval?

Comment 7

[Josh Matthews [:jdm]]

As long as the patch applies and builds, nobody will question how it was created. That being said, a patch against mozilla-central is certainly the easiest way forward. With regards to testing, we can throw it at a server to run them for us.

Comment 8

[rmkn85]

I'm porting part of the following commit from Chromium, will prepare a patch in a moment:

http://src.chromium.org/viewvc/chrome?view=rev&revision=84461
http://src.chromium.org/viewvc/chrome/trunk/src/ipc/ipc_channel_win.cc?r1=83362&r2=84461

Comment 9

[rmkn85]

Attachment: Remove GetLogonSessionOnlyDACL from ipc_channel_win

Porting commit from Chromium
http://src.chromium.org/viewvc/chrome?view=rev&revision=84461
http://src.chromium.org/viewvc/chrome/trunk/src/ipc/ipc_channel_win.cc?r1=83362&r2=84461

Comment 10

[Josh Matthews [:jdm]]

Pushed to our try server for testing: https://tbpl.mozilla.org/?tree=Try&rev=26f7da157aef

Comment 11

[rmkn85]

Did you run it against latest trunk source code, and there will be a build to download when it finishes?
Can you also make it build a patched XulRunner 14, so I can confirm it fixes the issue in my environment?

Comment 12

[Josh Matthews [:jdm]]

Sorry for the radio silence here. I've committed your patch in https://hg.mozilla.org/integration/mozilla-inbound/rev/546e2b61da57. Getting tryserver to produce a xulrunner build (even forgetting about the requirement for 14) is an uncertain process that I don't really have time to dig into at the moment, unfortunately.

Comment 13

[Josh Matthews [:jdm]]

However, the try link above does have Firefox builds available, if that is of any use to you.

Comment 14

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/546e2b61da57