Re: Sound devices under vservers?

Hye,

On Wed, 2016-09-07 at 18:57 +0200, Herbert Poetzl wrote:
> On Wed, Sep 07, 2016 at 11:33:48PM +1200, Andrew Ruthven wrote:
> > 
> > I run slimserver and squeezelite within a vserver, and until
> > recently they've worked nicely. But now squzeelite (which
> > actually plays the music out a soundcard) has stopped working.
> The files in /dev/snd within the container still match what is
> > on the hypervisor.
> 
> There is no hypervisor in Linux-VServer.
> There is a kernel and a host system and a number of guests.

Yeah, I was just stealing the language from elsewhere. ;)

> > 
> > Running strace when I start squeezelite gives me:
> 
> > 
> > open("/dev/snd/controlC0", O_RDONLY|O_CLOEXEC) = 4
> > fcntl(4, F_SETFD, FD_CLOEXEC)           = 0
> > ioctl(4, SNDRV_CTL_IOCTL_CARD_INFO or UI_DEV_CREATE,
> > 0x7ffcc50789d0) =
> > -1 ENOTTY (Inappropriate ioctl for device)
> 
> Try to run squeezelite with strace in a chroot into the guest
> and see if the result differs. 
> 
> If you get the same result, then the isolation (Linux-VServer 
> is not involved). 
> 
> If it works inside the chroot but fails inside the guest, then
> you might have insufficient capabilities.

Okay, fails within a chroot as well. Relevant lines:

From the host (setting LD_LIBRARY_PATH to pick up the libraries within
the guest):

open("/dev/snd/controlC0", O_RDONLY|O_CLOEXEC) = 4
fcntl(4, F_SETFD, FD_CLOEXEC)           = 0
ioctl(4, SNDRV_CTL_IOCTL_CARD_INFO or UI_DEV_CREATE, 0x7ffc646fd6d0) =
0
close(4)                                = 0

Within a chroot:

open("/dev/snd/controlC0", O_RDONLY|O_CLOEXEC) = 4
fcntl(4, F_SETFD, FD_CLOEXEC)           = 0
ioctl(4, SNDRV_CTL_IOCTL_CARD_INFO or UI_DEV_CREATE, 0x7ffd93373400) =
-1 ENOTTY (Inappropriate ioctl for device)
close(4)                                = 0

Within the guest:

open("/dev/snd/controlC0", O_RDONLY|O_CLOEXEC) = 4
fcntl(4, F_SETFD, FD_CLOEXEC)           = 0
ioctl(4, SNDRV_CTL_IOCTL_CARD_INFO or UI_DEV_CREATE, 0x7ffd6119bd40) =
-1 ENOTTY (Inappropriate ioctl for device)
close(4)

Aw crap. Turns out the minor numbers were different for
/dev/snd/controlC0 between the host and the guest. I had initially
looked at those and thought they matched, guess I read them wrong.

I guess I should add a step to my start script for the guest to always
copy over the contents of /dev/snd.  ;)  Actually, now it's a bind
mount.

Sorry to distract folks, but I thought I'd respond with the details and
the cause in case other folks hit it.

Cheers,
Andrew

-- 
Andrew Ruthven, Wellington, New Zealand
MIITP, ITCP

At work: [email protected]
At home: [email protected]
Cloud  : NZs only real cloud - https://catalyst.net.nz/cloud
GPG fpr: C603 FC4E 600F 1CEC D1C8  D97C 4B53 D931 E4D3 E863
LCA2017: The Future of Open Source, Hobart, AU - http://linux.conf.au