On 5/22/24 01:19, Eric Wheeler wrote:
On Sun, 19 May 2024, Hannes Reinecke wrote:
On 5/18/24 00:04, Eric Wheeler wrote:
On Fri, 17 May 2024, Hannes Reinecke wrote:
On 5/17/24 02:18, Eric Wheeler wrote:
Hello everyone,
Is there any work being done on namespaces for device-mapper targets, or
for the block layer in general?
For example, namespaces could make `dmsetup table` or `losetup -a` see
only devices mapped in that name space. I found this article from to 2013,
but it is quite old:
https://lwn.net/Articles/564854/
If you know any more recent work on the topic that I would be interested.
Thank you for help!
It is on my to-do list.
We sure should work on that one.
How you envision hooking namespaces into the block layer?
Overall idea is to inherit devices from the original namespace.
- upon creation the new namespace inherits all devices from the
original ns.
For namespace initialization, is there way to start with an empty
namespace (no inherit), and only add devices the namespace that you would
like to provide to the container? For example, you might want to provide a
logical volume to the container and then let the container users do with
they want in terms of creating new devices from that namespace-assigned
"root level" device.
Somehow it needs to be safe in terms of the container users changing the
device mapper table spec of a "root level" device using `dmsetup reload
--table`.
... except that you can't add anything as you won't have a tty, and
hence can't start a shell. And you might not be able to call 'malloc',
as glibc cannot call mmap() on /dev/zero.
And the plan is to be introduce namespaces for block devices, not for
character devices, so all character devices need to show up in all
namespaces.
Cheers,
Hannes
--
Dr. Hannes Reinecke Kernel Storage Architect
hare@xxxxxxx +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
