LWN: Comments on "Private loop devices with loopfs"

Private loop devices with loopfs

hallyn — Tue, 04 Oct 2022 18:50:06 +0000

(years late, but referencing even longer timelines)

Indeed Cellrox implemented device namespaces for android phones which, for instance, virtualized the display so that at any time one android container would have the real display while the other had a null display. See https://lwn.net/Articles/564854/ for instance from 2013 :)

Private loop devices with loopfs

Kamilion — Fri, 15 May 2020 19:37:59 +0000

Is it possible to loopback mount a file that grows yet?
I still can't believe supporting the DMG-like behavior of OSX is so difficult.

Private loop devices with loopfs

rwmj — Thu, 14 May 2020 10:28:12 +0000

This problem of separate "/dev spaces" must arise in other contexts too surely. TTYs? Non-loop block devices? I'm surprised that this problem hasn't already been solved more generically for any /dev subset.

Private loop devices with loopfs

samuelkarp — Thu, 14 May 2020 04:35:20 +0000

Docker's devicemapper-based storage driver can be backed by a loopback device.

Private loop devices with loopfs

jezuch — Tue, 12 May 2020 05:36:32 +0000

It may be a silly question, seeing how nobody mentioned it before, which probably means this is very obvious, but why not a loop device namespace? Wouldn't it be more consistent with the rest of the system?

Private loop devices with loopfs

ncultra — Mon, 11 May 2020 16:29:22 +0000

I feel Brauner's response to Hellwig's un-artful and wrong criticism was positive, to-the-point, and convincing. It is a great example of how to deal with what is essentially bullying on lkml or its component lists. I also agree with Brauner that this patch set fulfills more than one needed use case.

Private loop devices with loopfs

maxfragg — Mon, 11 May 2020 06:53:53 +0000

My answer would rather be a "no because of security/availability", by design FUSE allows to block the Kernel, which is a good reason not to use it in any true multi-user env.

Hidden devices

simcop2387 — Sun, 10 May 2020 22:20:52 +0000

A quick glance didn't seem to indicate it right away, but does anyone know if this mode works with raw disk images? in particular ones that are partitioned (GPT, MBR, etc). With the normal loopback devices you can tell the kernel to read the partition table from it (losetup -P in particular) and then use that to mount a given partition inside the image without having to do any offset, surgery, or other calculations to do it manually

Private loop devices with loopfs

edomaur — Sat, 09 May 2020 06:01:38 +0000

Multi user, probably as in "each layer of the fs can be freely read from any running container without security problems" ?

Private loop devices with loopfs

dtlin — Fri, 08 May 2020 19:07:41 +0000

https://www.sourceware.org/lvm2/wiki/DMLoop existed but doesn't look like it was completed or upstreamed.

Private loop devices with loopfs

Sesse — Fri, 08 May 2020 14:36:34 +0000

dm devices are virtual block devices where the blocks are mapped to other block devices (well, not necessarily directly; they typically go through a module of some sort to add logic). Loopback devices are virtual block devices where the blocks are mapped to a file; think mounting an ISO file as if it were a CD-ROM.

I guess in theory, you could create a loop backend for dm? But they are separate.

Private loop devices with loopfs

corbet — Fri, 08 May 2020 13:47:54 +0000

Loop devices predate the device mapper by a long time and are a separate thing.

Private loop devices with loopfs

theonewolf — Fri, 08 May 2020 12:17:38 +0000

Could you elaborate by what you mean with "multi user" and "performant"?

Also, do you have a link describing the seL4 work?

Private loop devices with loopfs

scientes — Fri, 08 May 2020 08:41:39 +0000

Nobody has managed to make a performant secure multi-user filesystem yet. seL4 people made a first attempt which was not successful.

Private loop devices with loopfs

pabs — Fri, 08 May 2020 04:44:35 +0000

Are loop devices part of the device mapper infrastructure or separate to it?

Private loop devices with loopfs

NYKevin — Thu, 07 May 2020 19:35:58 +0000

The Chromium OS docs have this line:

> If you have a use case that wouldn't be solved by supporting FUSE, please file a bug for us.

This does somewhat raise the question of whether FUSE is already an adequate replacement for loop devices in the context of containers. I imagine the answer is "No, because performance," but I would be very interested in seeing actual load tests and experimental results, if any exist.

Hidden devices

corbet — Thu, 07 May 2020 18:19:54 +0000

There is a provision for creating detached mounts directly with the new API, yes. We sort of skipped over that in the article, but probably should not have. The changelog to this patch describes it briefly.

Private loop devices with loopfs

josh — Thu, 07 May 2020 18:12:51 +0000

Legacy applications aside, I wonder if loop-control could just have an ioctl that returns a new file descriptor for a loopback device, and then the new mount API could support passing that file descriptor in to mount it, without having to give it a "loopN" device name.