Last updated:
0 purchases
linuxnsrel 1.0.6
Linux Kernel Namespace Relations
NOTE: Linux Kernel 4.9+ and Python 3.5+ supported only
This Python 3 package allows discovering the following Linux Kernel
namespace relationships and properties, without having to delve into
ioctl() hell:
the owning user namespace of another Linux kernel namespace.
the parent namespace of either a user or a PID namespace.
type of a Linux kernel namespace: user, PID, network, ...
owner user ID of a user namespace.
See also ioctl() operations for Linux namespaces
for more background information of the namespace operations exposed by
this Python library.
Installation
$ pip3 install linuxns-rel
NOTE: the installation automatically detects Debian 9 Stretch and
correctly handles installing the PyQt5 package, if it is missing, in
at most version 5.10.1 -- as this is the last known working PyQt5
release on Debian 9 Stretch.
API Documentation
Please head over to our
linuxns_rel API documentation
on GitHub Pages.
CLI Examples
NOTE: Debian 9 Stretch users will lack the ability discover the
owner user ID of hierarchical namespaces. The reason is that Debian
9 Stretch ships with outdated Linux kernels that lack the required
Kernel functionality. Please enjoy this kind of stability. (Or is
this "austerity"?)
List User Namespaces
You can either use a simple lsuserns or sudo lsuserns -- the latter
ensures that you'll see all user namespaces instead of only those you have
user access to. Optionally, you can also see which other namespaces are owned
by specific user namespaces.
NOTE: lsuserns (and lspidns) discover user and pid namespaces from
more locations than lsns does:
/proc/$PID/ns/ (the only location supported by lsns)
/proc/$PID/fd/ (open file descriptors referencing namespaces; only
lsuserns atm)
namespace ioctl() (see "man 2 ioctl_ns")
$ sudo lsuserns
may yield something like this, a pretty hierarchy of Linux kernel user
namespaces:
user:[4026531837] process "init (1)" namespace-owning user "root" (0)
├── user:[4026532217] process "sleep (9403)" namespace-owning user "harald" (1000)
├── user:[4026532401] process "firefox (1374)" namespace-owning user "harald" (1000)
├── user:[4026532582] process namespace-owning user "harald" (1000)
│ └── user:[4026532583] process namespace-owning user "harald" (1000)
│ └── user:[4026532584] process namespace-owning user "harald" (1000)
│ └── user:[4026532585] process "bash (1468)" namespace-owning user "harald" (1000)
├── user:[4026532639] process "firefox (2723)" namespace-owning user "harald" (1000)
├── user:[4026532696] process "firefox (2768)" namespace-owning user "harald" (1000)
└── user:[4026532936] process "firefox (17192)" namespace-owning user "harald" (1000)
If you want to additionally see which non-user namespaces are owned by the user namespaces, then add the CLI arg -d (as in "details"):
$ sudo lsuserns -d
NOTE: -c colorizes the output, differentiating namespaces in color
based on their types.
In the output, "⟜" denotes a non-user namespace (ipc, mnt, net, ...) which is
owned by a particular user namespace.
user:[4026531837] process "init (1)" namespace-owning user "root" (0)
├── ⟜ cgroup:[4026531835] process "init (1)"
├── ⟜ ipc:[4026531839] process "init (1)"
├── ⟜ mnt:[4026531840] process "init (1)"
├── ⟜ mnt:[4026531860] process "[kdevtmpfs] (57)"
├── ⟜ mnt:[4026532216] process "systemd-udevd (470)"
├── ⟜ mnt:[4026532263] process "systemd-resolved (1217)"
├── ⟜ mnt:[4026532264] process "systemd-timesyncd (1218)"
├── ⟜ mnt:[4026532452] process "haveged (1223)"
├── ⟜ mnt:[4026532453] process "ModemManager (1264)"
├── ⟜ mnt:[4026532454] process "NetworkManager (1268)"
├── ⟜ mnt:[4026532455] process "irqbalance (1312)"
├── ⟜ mnt:[4026532456] process "mpd (1402)"
├── ⟜ mnt:[4026532524] process "upowerd (1618)"
├── ⟜ net:[4026532000] process "init (1)"
├── ⟜ net:[4026532266] process "haveged (1223)"
├── ⟜ net:[4026532458] process "rtkit-daemon (2148)"
├── ⟜ pid:[4026531836] process "init (1)"
├── ⟜ uts:[4026531838] process "init (1)"
├── user:[4026532217] process "sleep (9403)" namespace-owning user "harald" (1000)
│ └── ⟜ net:[4026532219] process "sleep (9403)"
├── user:[4026532401] process "firefox (1374)" namespace-owning user "harald" (1000)
│ ├── ⟜ ipc:[4026532512] process "firefox (1374)"
│ └── ⟜ net:[4026532529] process "firefox (1374)"
├── user:[4026532582] process namespace-owning user "harald" (1000)
│ └── user:[4026532583] process namespace-owning user "harald" (1000)
│ └── user:[4026532584] process namespace-owning user "harald" (1000)
│ └── user:[4026532585] process "bash (1468)" namespace-owning user "harald" (1000)
├── user:[4026532639] process "firefox (2723)" namespace-owning user "harald" (1000)
│ ├── ⟜ ipc:[4026532640] process "firefox (2723)"
│ └── ⟜ net:[4026532642] process "firefox (2723)"
├── user:[4026532696] process "firefox (2768)" namespace-owning user "harald" (1000)
│ ├── ⟜ ipc:[4026532697] process "firefox (2768)"
│ └── ⟜ net:[4026532699] process "firefox (2768)"
└── user:[4026532936] process "firefox (17192)" namespace-owning user "harald" (1000)
├── ⟜ ipc:[4026532937] process "firefox (17192)"
└── ⟜ net:[4026532939] process "firefox (17192)"
If you have either Chromium or/and Firefox running, then these will
add some user namespaces in order to sandbox their inner workings. And
to add in some more hierarchical user namespaces, in another terminal
session simply issue the following command:
$ unshare -Ur unshare -Ur unshare -Ur unshare -Ur
Debian users may need to sudo because their distro's default
configuration prohibits ordinary users to create new user namespaces.
List PID Namespaces
$ sudo lspidns
shows the PID namespace hierarchy, such as:
pid:[4026531836] process "init" owner user:[4026531837] root (0)
├── pid:[4026532532] process "sh" owner user:[4026531837] root (0)
└── pid:[4026532468] process "chromium-browser --type=zygote" owner user:[4026532467] foobar (1000)
├── pid:[4026532464] process "chromium-browser" owner user:[4026532589] foobar (1000)
├── pid:[4026532466] process "chromium-browser" owner user:[4026532589] foobar (1000)
└── pid:[4026532590] process "chromium-browser" owner user:[4026532589] foobar (1000)
Don't worry that the PID namespace hierarchy doesn't match the user
namespace hierarchy. That's perfectly fine, depending on which programs
run. In our example, we didn't create new PID namespaces when using
unshare, so we see only additional PID namespaces created by
Chromium (Firefox doesn't create them though).
Namespace Graph
Okay, with the famous graphviz installed (apt-get install graphviz)
now simply do:
$ sudo -E graphns
Note: -E ensures that the graph viewer will correctly use the your
desktop environment theme.
...and you get something fancy with arrows, et cetera, in a cute (erm,
"qute") little viewer window:
The view window supports these actions:
mouse drag: move the visible area around inside the view window
(usefull when either the graph is really large or when zoomed in).
mouse wheel: zoom in/out.
+/- keys: zoom in/out.
1 key: reset zoom to 1x.
s key: save image to an SVG file.
q key: close and exit the viewer.
h key: show help on mouse and keys.
Potentially FAQs
Q: Why do get_userns() and get_parentns() return file objects
(TextIO) instead of filesystem paths?
A: Because that's what the Linux namespace-related ioctl()
functions are giving us: open file descriptors referencing namespaces
in the special nsfs namespace filesystem. There are no paths
associated with them.
Q: What argument types do get_nstype(), get_userns(),
get_parentns(), and get_owner_uid() expect?
A: Choose your weapon:
a filesystem path (name), such as /proc/self/ns/user,
an open file object (TextIO), such as returned by open(),
an open file descriptor, such as returned by fileno() methods.
Q: Why does get_parentns() throw an PermissionError?
A: There are multiple causes:
you didn't specify a PID or user namespace,
the parent namespace either doesn't exist,
or the parent namespace is inaccessible to you,
oh, you really have no access to the namespace reference.
Q: Why does get_userns() throw an PermissionError?
A: You don't have access to the owning user namespace.
For personal and professional use. You cannot resell or redistribute these repositories in their original state.
There are no reviews.