tests/procdesc: Add some test cases for pdopenpid()

[kib]

i will check this against my branch later. Could you please mail me the git-format patch?

tests/sys/kern/procdesc.c

257

It would be more holistic to loop around the pause()

355

[markj]

Review feedback, fix the capmode test, add a new test which triggers EMFILE.

[markj]

Take kib's patch for the self test

[kib]

Do we want some test to ensure that pdopenpid() has no effect on the target? I mean, ptrace(PT_ATTACH) might cause spurious wakeups (partially mitigated now).

tests/sys/kern/procdesc.c

228

I intended that pause() is covered by loop in all tests.

350

You are using sleep() there to eventually exit the child, am I correct?
You should still loop around the sleep(), otherwise the test is flaky. In the loop, it is possible to check the parent pid with getppid() to note that the parent changed because original parent exited, and then exit the child.

[kib]

tests/sys/kern/procdesc.c

350

diff --git a/tests/sys/kern/procdesc.c b/tests/sys/kern/procdesc.c
index 6258ecd6e8d8..2f54f138f12e 100644
--- a/tests/sys/kern/procdesc.c
+++ b/tests/sys/kern/procdesc.c
@@ -342,12 +342,14 @@ ATF_TC_BODY(pdopenpid_esrch, tc)
 ATF_TC_WITHOUT_HEAD(pdopenpid_capmode);
 ATF_TC_BODY(pdopenpid_capmode, tc)
 {
- pid_t child;
+  pid_t child, parent;
 
+  parent = getpid();
   child = fork();
   ATF_REQUIRE_MSG(child >= 0, "fork: %s", strerror(errno));
   if (child == 0) {
-    sleep(1);
+     while (getppid() == parent)
+        sleep(1);
      _exit(0);
   }

worked for me

[markj]

Address review feedback, add procdesc_no_wakeup

[markj]

In D58023#1331065, @kib wrote:

Do we want some test to ensure that pdopenpid() has no effect on the target? I mean, ptrace(PT_ATTACH) might cause spurious wakeups (partially mitigated now).

Like, we should check that pdopenpid() does not result in a spurious wakeup of the target? That's a good idea.

tests/sys/kern/procdesc.c

350

Right, since the parent is in capability mode here it can't use kill().

[kib]

tests/sys/kern/procdesc.c

299

Loop there?

651

This is racy, and I think that it would typically cause the test to succeed even for the interrupting pdopenpid(). The child might not yet exit.

The proper test would be to re-check that the child is still sleeping in "piperd" after pdopen().

[markj]

Re-check that the child is sleeping.