[LTP] [PATCH] ioctl_pidfd02-06: Add CONFIG_USER_NS and CONFIG_PID_NS to needs_kconfigs

[Cyril Hrubis]

Hi!
> > > Because that was my question - really always prefer kconfig even there is a
> > > simple runtime solution? I'd like to have some "rule" like conclusion we can
> > > point during review.
> >
> > I think that from a long term view this is going to be simpler solution
> > than having many different types of checks. The less diverse these
> > checks are the easier they are to review and maintain. Hence I lean
> > towards kernel config checks even though they are slower (mostly
> > unmeasurable on today's harware) than the alternatives.
> 
> I think I lean opposite way, and rather have a check for right
> environment to support the test.
> You can have feature X enabled in kernel config, but still disabled
> later at boot/runtime
> (e.g. max_user_namespaces=0), or a module simply not being loaded.

That is a good catch.

Maybe the best way forward would be to add hooks for certain config
options into the LTP kernel config parser that would do additional
runtime checks. That way we would have both the information on which
kernel configs should be enabled in test metadata as well as runtime
checks.

What about eventually adding something as:

diff --git a/lib/tst_kconfig.c b/lib/tst_kconfig.c
index 9bcd57721..f6abe6cc7 100644
--- a/lib/tst_kconfig.c
+++ b/lib/tst_kconfig.c
@@ -110,6 +110,18 @@ static void close_kconfig(FILE *fp)
                fclose(fp);
 }

+static void runtime_check(struct tst_kconfig_var *var)
+{
+       if (strstr(var->id, "CONFIG_USER_NS")) {
+               if (!tst_user_ns_enabled()) {
+                       tst_res(TINFO, "CONFIG_USER_NS present but runtime is disabled");
+                       var->val = 'n';
+               }
+       } else if (...)
+               ...
+       }
+}
+
 static inline int kconfig_parse_line(const char *line,
                                      struct tst_kconfig_var *vars,
                                      unsigned int vars_len)
@@ -183,9 +195,11 @@ out:
                        switch (val[0]) {
                        case 'y':
                                vars[i].choice = 'y';
+                               runtime_check(&vars[i]);
                                return 1;
                        case 'm':
                                vars[i].choice = 'm';
+                               runtime_check(&vars[i]);
                                return 1;
                        }
                }


-- 
Cyril Hrubis
chrubis@suse.cz