[RFC PATCH 0/8] namespace-aware configfs
Hannes Reinecke
hare at kernel.org
Sat Jun 13 04:14:29 PDT 2026
Hey all,
during discussions at LSF/MM I thought it would be quite helpful
to make configfs namespace aware, such that one could do:
unshare --mnt
mount -t configfs none /sys/kernel/config
and get a new instance of configfs. This will allow to have distinct
configurations for each namespace.
This is particularly helpful if you want to run eg nvmet in a container;
with this patchset each container can have its own configuration, and
one can simulate a multi-node nvmet cluster using containers.
How it works:
The configfs 'root' entries are stored in 'configfs_super_info' which
is stored in an xarray using the namespace ID as the index.
A configfs subsystem can be converted to be namespace aware by implementing
the new subsyste callbacks 'fill_subsystem()' and 'clear_subsystem', which
are responsible for populating the subsystem structure.
Upon registration via 'register_subsystem()' the subsystem is put in a
linked list within 'configfs_super_info'. So when mount() is called from
a different namespace this list is traversed, and new subsystems are created
in the new namespace using the callbacks.
Open Issues:
- I've added a new function 'mnt_clone_direct()' to clone the vfsmount
entry (the original code just did a simple_pin_fs()). Not sure if
that's correct. Christian?
- The current cloning mechanism is not really a hierarchy, but rather
always using the default namespace to find registered subsystems.
Meaning you can only call 'register_subsystem()' for the default
namespace. But then one shouldn't call modprobe in a container, so
that's okay I guess.
- The original content of the configfs remains visible even from
within the container, and the new 'mount' will just overlay that.
Ideally I would have the container start off with an empty
/sys/kernel/config to avoid configuration issues. But again I've
no idea how to do that (or if it's even possible).
Otherwise, as usual, commands and reviews are welcome.
Hannes Reinecke (8):
fs/configfs: rework configfs_is_root()
fs/configfs: dynamically allocate super_info
fs/configfs: separate out configfs_{link,unlink}_root()
fs/namespace: implement mnt_clone_direct()
fs/configfs: add superblock as attribute to configfs_pin_fs()
fs/configfs: add 'fill_subsystem' and 'clear_subsystem' callbacks
fs/configfs: switch to get_tree_keyed()
nvmet: make configfs setup namespace aware
drivers/nvme/target/configfs.c | 183 ++++++++++++++++++++++-----
drivers/nvme/target/core.c | 35 +++---
drivers/nvme/target/discovery.c | 101 +++++++++++----
drivers/nvme/target/nvmet.h | 13 +-
fs/configfs/configfs_internal.h | 21 +++-
fs/configfs/dir.c | 167 +++++++++++++++++++-----
fs/configfs/mount.c | 216 +++++++++++++++++++++++++++-----
fs/namespace.c | 11 ++
include/linux/configfs.h | 7 ++
include/linux/mount.h | 1 +
11 files changed, 615 insertions(+), 144 deletions(-)
--
2.51.0
More information about the Linux-nvme
mailing list