FS#36259 : [systemd] 206 sets incorrect permissions on static nodes

FS#36259 - [systemd] 206 sets incorrect permissions on static nodes

Attached to Project: Arch Linux
Opened by Gaetan Bisson (vesath) - Wednesday, 24 July 2013, 02:50 GMT
Last edited by Dave Reisner (falconindy) - Tuesday, 13 August 2013, 16:45 GMT

Task Type	Bug Report
Category	Packages: Testing
Status	Closed
Assigned To	Dave Reisner (falconindy) Tom Gundersen (tomegun)
Architecture	All
Severity	Medium
Priority	Normal
Reported Version
Due in Version	Undecided
Due Date	Undecided
Percent Complete
Votes	7 Thomas Lucke (Lucky) (2013-08-01) Marti (intgr) (2013-07-29) Kohei Yoshitomi (anergy) (2013-07-29) Anonymous Submitter (2013-07-28) Kerrick Staley (KerrickStaley) (2013-07-28) Christian Hesse (eworm) (2013-07-26) Giuliano Schneider (gs93) (2013-07-24)
Private	No

Details

With linux-3.10.2-1 from [testing] the fuse module is not automatically loaded when a client application tries to use it - as used to be the case with linux-3.9.9-1 and before.

This task depends upon

Closed by Dave Reisner (falconindy)
Tuesday, 13 August 2013, 16:45 GMT
Reason for closing: Fixed
Additional comments about closing: systemd-206-2

Comment by Gaetan Bisson (vesath) - Wednesday, 24 July 2013, 02:51 GMT

That's all I can say for now; when I find time to investigate further I'll report my findings here if any.

Comment by Gerardo Exequiel Pozzi (djgera) - Wednesday, 24 July 2013, 03:52 GMT

maybe related to recent udev/kmod changes? /dev/fuse previoulsy has 0666 now has 0600

Comment by Tom Gundersen (tomegun) - Wednesday, 24 July 2013, 11:07 GMT

50-udev-default.rules says that /dev/fuse should be 0666.

So this seems to be an udev bug (almost certainly introduced by me). Sadly I don't have time to look into it today.

Comment by Tom Gundersen (tomegun) - Wednesday, 24 July 2013, 11:29 GMT

Hm, my OCD won, so I had a look. I can not reliably reproduce this... If you manually run "/lib/systemd/systemd-udevd --debug" you should see calls to "chmod" near the end of the output (as long as the modules are not loaded). What happens there?

The relevant code is http://cgit.freedesktop.org/systemd/systemd/tree/src/udev/udev-rules.c#n2500

I'll have a better look tomorrow if this is not resolved by then.

Comment by Dave Reisner (falconindy) - Wednesday, 24 July 2013, 12:34 GMT

I can reproduce this on a VM -- it boots with 0600 permissions, but as soon as I load the fuse module, udev changes the perms to 0666.

Comment by Tom Gundersen (tomegun) - Wednesday, 24 July 2013, 12:41 GMT

@falconindy: that's my guess, but I can't make much sense of it yet.

I can reproduce the problem too now, but if I restart udevd it fixes itself, so it looks like a race-condition.

I first thought it may be that the device nodes are somehow not created before udev is started, but that doesn't make sense as both the GROUP= and the ACL stuff is applied correctly, it is just the permissions that are wrong...

Comment by Tom Gundersen (tomegun) - Wednesday, 24 July 2013, 13:27 GMT

Problem was that the following happens:

systemd-tmpfiles --prefix=/dev --create <- createse dead nodes
systemd-udevd <- changes permissions
systemd-tmpfiles --create <- resets permissions

I'll post a fix to the systemd ML shortly, which will change the tmpfiles api slightly to allow:

systemd-tmpfiles --prefix=/dev --create <- createse dead nodes
systemd-udevd <- changes permissions
systemd-tmpfiles --prefix=!/dev --create <- don't touch /dev at all

Comment by Tom Gundersen (tomegun) - Wednesday, 24 July 2013, 13:35 GMT

Patch: http://lists.freedesktop.org/archives/systemd-devel/2013-July/012261.html

Comment by Dave Reisner (falconindy) - Wednesday, 24 July 2013, 16:07 GMT

Fixed by: http://cgit.freedesktop.org/systemd/systemd/commit/?id=ec99834cb0

But this is made possible through 2 prior commits:

http://cgit.freedesktop.org/systemd/systemd/commit/?id=a2aced4add
http://cgit.freedesktop.org/systemd/systemd/commit/?id=5c7951141f

Comment by Gaetan Bisson (vesath) - Wednesday, 24 July 2013, 23:41 GMT

Thanks Tom and Dave for figuring this out and fixing it so quickly!

Comment by Christian Hesse (eworm) - Friday, 26 July 2013, 08:36 GMT

Thanks for providing a fix for that. Will there be an updated systemd package in [testing] soon?

Comment by Dave Reisner (falconindy) - Friday, 26 July 2013, 14:04 GMT

No, I'd rather see a 207 release given that there's other problems with 206.