[LTP] [REGRESSION] Broken tests using tst_net.sh by 893ca0abe7 (was: [PATCH 1/2] lib: multiply the timeout if detect slow kconfigs)

[Petr Vorel]

Hi Li, Cyril, all,

...
> +++ b/lib/tst_test.c
> @@ -555,9 +555,6 @@ static int multiply_runtime(int max_runtime)

>  	parse_mul(&runtime_mul, "LTP_RUNTIME_MUL", 0.0099, 100);

> -	if (tst_has_slow_kconfig())
> -		max_runtime *= 4;
> -
>  	return max_runtime * runtime_mul;
>  }

> @@ -1706,6 +1703,9 @@ unsigned int tst_multiply_timeout(unsigned int timeout)
>  	if (timeout < 1)
>  		tst_brk(TBROK, "timeout must to be >= 1! (%d)", timeout);

> +	if (tst_has_slow_kconfig())
> +		timeout *= 4;

FYI this change, merged as 893ca0abe7 ("lib: multiply the timeout if detect slow
kconfigs") caused a regression on *all* tests which use tst_net.sh.

First, it detects slow config on everything which uses struct tst_test,
which are unfortunately some tools at testcases/lib/:

$ git grep -l "struct tst_test" testcases/lib/*.c
testcases/lib/tst_device.c // not obvious reason, might be removed
testcases/lib/tst_get_free_pids.c // force messages to be printed from new library
testcases/lib/tst_ns_create.c // .forks_child = 1, Needed by SAFE_CLONE
testcases/lib/tst_ns_exec.c  // .forks_child = 1, Needed by SAFE_CLONE
testcases/lib/tst_run_shell.c // not obvious reason, might be removed

Besides unimportant fact that slow config detection is an unnecessary slowdown
on these tools, the problem is that it prints messages to stderr, which causes
breakage.

Previously tst_ns_exec.c run just command passed by -c parameter:

$ ./tst_ns_exec 14536 net,mnt sh -c " cat /proc/sys/net/ipv6/conf/ltp_ns_veth1/disable_ipv6"
0

Now it prints TINFO:
$ ./tst_ns_exec 14536 net,mnt sh -c " cat /proc/sys/net/ipv6/conf/ltp_ns_veth1/disable_ipv6"
tst_kconfig.c:88: TINFO: Parsing kernel config '/proc/config.gz'
tst_kconfig.c:667: TINFO: CONFIG_LATENCYTOP kernel option detected which might slow the execution
0

tst_rhost_run() in tst_net.sh runs binary on remote host:

	output=$($rcmd "$sh_cmd" $out 2>&1 || echo 'RTERR')

Redirect stderr to stdout (2>&1) is likely needed for some tests. But
tst_rhost_run() output is often parsed including therefore certain tools in
testcases/lib/ have to print only expected output:

init_ltp_netspace()
{
	...
		pid="$(ROD tst_ns_create net,mnt)"
	...
	export LTP_NETNS="${LTP_NETNS:-tst_ns_exec $pid net,mnt}"

I will probably solve it by adding yet another parameter to tst_rhost_run(),
which ignores stderr and use it for tst_ns_{create,exec}. But maybe there is
another solution (or another problem).

Kind regards,
Petr