Re: [PATCH] selftests/mm: handle fork() failure in migration tests
From: Mike Rapoport
Date: Mon May 11 2026 - 13:22:37 EST
On Mon, May 11, 2026 at 02:47:38PM +0200, David Hildenbrand (Arm) wrote:
> On 5/11/26 11:34, Hu Song wrote:
> > From: Song Hu <husong@xxxxxxxxxx>
> >
> > When fork() fails and returns -1, the code falls into the else branch
> > and stores -1 into self->pids[i]. Later, kill(-1, SIGTERM) is called
> > which sends SIGTERM to every process the user has permission to signal,
> > potentially killing the entire user session.
> >
> > Add an explicit check for fork() returning -1 (pid < 0). On failure,
> > clean up any already-forked children before failing the test via
> > ASSERT_GE(pid, 0). This fix is applied to all three fork() call sites
> > in shared_anon, shared_anon_thp, and shared_anon_htlb tests.
> >
> > Signed-off-by: Hu Song <husong@xxxxxxxxxx>
> > ---
> > tools/testing/selftests/mm/migration.c | 15 +++++++++++++++
> > 1 file changed, 15 insertions(+)
> >
> > diff --git a/tools/testing/selftests/mm/migration.c b/tools/testing/selftests/mm/migration.c
> > index 60e78bbfc0e3..f433e4f195ad 100644
> > --- a/tools/testing/selftests/mm/migration.c
> > +++ b/tools/testing/selftests/mm/migration.c
> > @@ -163,6 +163,11 @@ TEST_F_TIMEOUT(migration, shared_anon, 2*RUNTIME)
> > memset(ptr, 0xde, TWOMEG);
> > for (i = 0; i < self->nthreads - 1; i++) {
> > pid = fork();
> > + if (pid < 0) {
> > + while (--i >= 0)
> > + kill(self->pids[i], SIGTERM);
>
>
>
> > + ASSERT_GE(pid, 0);
> > + }
> > if (!pid) {
>
> I think "else if" reads nicer here.
>
> But why do we care about cleaning up the other processes? IIUC, we don't do that
> explicitly if e.g.,
>
> ASSERT_EQ(migrate(ptr, self->n1, self->n2), 0);
>
> fails?
We care about cleaning up all the forked processes because otherwise they
remain zombies :)
But killing some of them if a fork() failed does not help.
I have a more structured fix here:
https://lore.kernel.org/all/20260511162840.375890-5-rppt@xxxxxxxxxx
>
> --
> Cheers,
>
> David
--
Sincerely yours,
Mike.