On Fri, Jun 26, 2020 at 03:45:04PM -0700, Andrii Nakryiko wrote:
> On Fri, Jun 26, 2020 at 10:56 AM Martin KaFai Lau <kafai@xxxxxx> wrote:
> >
> > It is common for networking tests creating its netns and making its own
> > setting under this new netns (e.g. changing tcp sysctl). If the test
> > forgot to restore to the original netns, it would affect the
> > result of other tests.
> >
> > This patch saves the original netns at the beginning and then restores it
> > after every test. Since the restore "setns()" is not expensive, it does it
> > on all tests without tracking if a test has created a new netns or not.
> >
> > Signed-off-by: Martin KaFai Lau <kafai@xxxxxx>
> > ---
> > tools/testing/selftests/bpf/test_progs.c | 21 +++++++++++++++++++++
> > tools/testing/selftests/bpf/test_progs.h | 2 ++
> > 2 files changed, 23 insertions(+)
> >
> > diff --git a/tools/testing/selftests/bpf/test_progs.c b/tools/testing/selftests/bpf/test_progs.c
> > index 54fa5fa688ce..b521ce366381 100644
> > --- a/tools/testing/selftests/bpf/test_progs.c
> > +++ b/tools/testing/selftests/bpf/test_progs.c
> > @@ -121,6 +121,24 @@ static void reset_affinity() {
> > }
> > }
> >
> > +static void save_netns(void)
> > +{
> > + env.saved_netns_fd = open("/proc/self/ns/net", O_RDONLY);
> > + if (env.saved_netns_fd == -1) {
> > + perror("open(/proc/self/ns/net)");
> > + exit(-1);
> > + }
> > +}
> > +
> > +static void restore_netns(void)
> > +{
> > + if (setns(env.saved_netns_fd, CLONE_NEWNET) == -1) {
> > + stdio_restore();
> > + perror("setns(CLONE_NEWNS)");
> > + exit(-1);
> > + }
> > +}
> > +
> > void test__end_subtest()
> > {
> > struct prog_test_def *test = env.test;
> > @@ -643,6 +661,7 @@ int main(int argc, char **argv)
> > return -1;
> > }
> >
> > + save_netns();
>
> you should probably do this also after each sub-test in test__end_subtest()?
You mean restore_netns()?
It is a tough call.
Some tests may only want to create a netns at the beginning for all the subtests
to use (e.g. sk_assign.c). restore_netns() after each subtest may catch
tester in surprise that the netns is not in its full control while its
own test is running.
I think an individual test should have managed the netns properly within its
subtests already if it wants a correct test result. It can unshare at the
beginning of each subtest to get a clean state (e.g. in patch 8).
test_progs.c only ensures a config made by an earlier test does
not affect the following tests.
