Tests fail without CONFIG_USER_NS=y

[dezgeg]

The testsuite fails on Arch Linux, presumablu due to lack of user namespaces:

error: cloning builder process: Invalid argument
error: unable to start build process
+ outPath=
FAIL: tests/linux-sandbox.sh

and

error: cloning builder process: Invalid argument
error: build of ‘/tmp/nix-build-nix-1.12pre5511_c94f3d55.drv-0/nix-test/store/qq9v4bnc71y93qpwy18vcm1k2kqylc1w-build-hook-input-1.drv’ on ‘local?root=/tmp/nix-build-nix-1.12pre5511_c94f3d55.drv-0/nix-test/store1’ failed: unable to start build process
cannot build derivation ‘/tmp/nix-build-nix-1.12pre5511_c94f3d55.drv-0/nix-test/store/8mdh71d0mw0y4awqmzmxkf595di4kfza-build-hook.drv’: 1 dependencies couldn't be built
error: build of ‘/tmp/nix-build-nix-1.12pre5511_c94f3d55.drv-0/nix-test/store/8mdh71d0mw0y4awqmzmxkf595di4kfza-build-hook.drv’ failed
+ outPath=
FAIL: tests/build-remote.sh

Presumably those tests could try calling some trivial unshare command and skip those tests if they don't work (except unshare comes from util-linux, not coreutils, sigh...).

[jcumming]

Also fails on SLES11-SP4.

[edolstra]

Does /proc/self/ns/user exist if CONFIG_USER_NS is disabled? If not, its existence can be used to decide whether to run the test.

[dezgeg]

It doesn't but unfortunately that most likely doesn't fix the Debian problem: #1625 (comment)

[dezgeg]

But I guess you could check the existence and/or value of /proc/sys/kernel/unprivileged_userns_clone...

[dezgeg]

#1797 should fix this using that method.