chromium: child processes escape the network namespace sandbox

[ghost]

Write clear, concise and in textual form.

Bug and expected behavior

• Describe the bug.

I am running Chromium in a custom netns:
firejail --netns=for_wg chromium --user-data-dir="/home/lockywolf/.config/chromium" --disable-async-dns
Then I am finding firejail's pid:
pgrep firejail => 31669 (I actually do this by looking at the process tree, as there are two firejails.)
Then I look at the process tree:

lockywolf@delllaptop:~/BACKUP$ pstree -pST 31669 
firejail(31669)───firejail(31670)───chromium(31695)─┬─chromium(428)
                                                    ├─chromium(31707)───chromium(31730)───chromium(31743)
                                                    ├─chromium(31708,net,pid,user)───chromium(31710)─┬─chromium(1092,pid)
                                                    │                                                ├─chromium(1190,pid)
                                                    │                                                ├─chromium(1478,pid)
                                                    │                                                ├─chromium(31754,pid)
                                                    │                                                ├─chromium(31770,pid)
                                                    │                                                ├─chromium(31780,pid)
                                                    │                                                ├─chromium(31792,pid)
                                                    │                                                ├─chromium(31799,pid)
                                                    │                                                ├─chromium(31808,pid)
                                                    │                                                ├─chromium(31818,pid)
                                                    │                                                ├─chromium(31824,pid)
                                                    │                                                ├─chromium(31841,pid)
                                                    │                                                ├─chromium(31859,pid)
                                                    │                                                ├─chromium(31909,pid)
                                                    │                                                ├─chromium(31924,pid)
                                                    │                                                ├─chromium(31935,pid)
                                                    │                                                ├─chromium(31936,pid)
                                                    │                                                ├─chromium(31976,pid)
                                                    │                                                └─chromium(32313,pid)
                                                    └─chromium(31735)

Huh? PID 31708 escaped the sandbox? Checking:

lockywolf@delllaptop:~/BACKUP$ sudo /sbin/ip netns identify 31708

lockywolf@delllaptop:~/BACKUP$

Nothing!

• What did you expect to happen?

I expected all Chromium's children to be in the for_wg netns.

No profile and disabling firejail

• What changed calling firejail --noprofile /path/to/program in a terminal?
  Nothing, same behaviour.
• What changed calling the program by path (check which <program> or firejail --list while the sandbox is running)?
  I do not understand the question.

Reproduce

• Steps to reproduce the behavior:
  See above.

Environment

• Linux distribution and version (ie output of lsb_release -a, screenfetch or cat /etc/os-release)
  Slackware 15.0 alpha1, kernel 5.10.21, glibc 2.33, chromium 89.0.4389.72 (Developer Build) (64-bit)
• Firejail version (output of firejail --version) exclusive or used git commit (git rev-parse HEAD)
  0.9.64.4

Additional context
Other context about the problem like related errors to understand the problem.
Exactly the same behaviour happens if I run chromium with ip netns exec chromium ...
Checklist

• The profile (and redirect profile if exists) hasn't already been fixed upstream.: I see nothing related to netns there.
• The program has a profile. (If not, request one in https://github.com/netblue30/firejail/issues/1139)
• I have performed a short search for similar issues (to avoid opening a duplicate).: Yes
• If it is a AppImage, --profile=PROFILENAME is used to set the right profile. : Not an appimage.
• Used LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8 PROGRAM to get english error-messages.
• I'm aware of browser-allow-drm yes/browser-disable-u2f no in firejail.config to allow DRM/U2F in browsers.: yes, but I do not use U2F and do not keep chrome executables in ~/

debug output

attaching stdout and stderr

firejail.chrome.debug.stderr.txt
firejail.chrome.debug.stdout.txt

The result of children being in a different namespace results in that different tabs of the browser are getting different routes and dns servers.

[rusty-snake]

I guess that's because of chromiums sandbox. What happens if you start chromium with --no-sandbox?

EDIT: Does slackwares kernel support unprivileged userns clone? (sysctl kernel.unprivileged_userns_clone)

[ghost]

What happens if you start chromium with --no-sandbox?

Does what I expect it to do. Every child is in for_wg. I would have closed this issue, but em... relying on chromium's good will doesn't infuse me with certainty.

Does slackwares kernel support unprivileged userns clone? (sysctl kernel.unprivileged_userns_clone)

root@delllaptop:/home/lockywolf# sysctl kernel.unprivileged_userns_clone
sysctl: cannot stat /proc/sys/kernel/unprivileged_userns_clone: No such file or directory

[rusty-snake]

Does slackwares kernel support unprivileged userns clone? (sysctl kernel.unprivileged_userns_clone)

root@delllaptop:/home/lockywolf# sysctl kernel.unprivileged_userns_clone
sysctl: cannot stat /proc/sys/kernel/unprivileged_userns_clone: No such file or directory

This sysctl is a debian patch (used by debian and arch). Mainline does not have it. So unprivileged_userns_clone is supported on your system if userns is supported at all (sysctl user/max_user_namespaces != 0). Then it should work for you the enable chromium-common-hardened.inc (firejail >= 0.9.65 chromium-common-hardened.profile):

mkdir -p ~/.config/firejail
echo "include chromium-common-hardened.inc" >> ~/.config/firejail/chromium-common.local

relying on chromium's good will doesn't infuse me with certainty.

Chromium has it's own sandbox (surprise 😎 ) with a suid-helper in /usr/lib64/chromium/chrome-sandbox (or somewhere else, depending on distro/packages/chromium,vivaldi,brave,..). Chromium fails to start on systems w/o unprivileged_userns_clone if it's started with NO_NEW_PRIVS=1, in a own userns, with seccomp filters or a capability bounding set w/o CAP_SYS_CHROOT and CAP_SYS_ADMIN. Therefore all these this are allowed/disable by default.

So it can run code as root in the default users with NO_NEW_PRIVS=0 and CAP_SYS_ADMIN. In general that's are good starting point to:

• setup new namespaces with less restrictions
• escape where ever it wants (if the suid binary on your systems allows it)
• undo blacklist and whitelist AFAICTY

You can also sudo chmod -s /usr/lib64/chromium-browser-privacy/chrome-sandbox if unprivileged_userns_clone is enabled. I do this via package-manager hocks (needs to be done after every update).

[rusty-snake]

Any progress here?

[rusty-snake]

I'm closing here due to inactivity, please fell free to request to reopen if you have more questions.