rdma_link_test.go (1)

117-160: ⚠️ Potential issue | 🟠 Major

firstRdmaDevice called after state mutation leaks newns and RDMA mode — likely cause of the CI failure.

By line 144, two mutations are already in effect:

1. RDMA system mode has been set to "exclusive" (line 126).
2. newns has been created (line 137) but is not deferred for close.

If firstRdmaDevice(t) calls t.Skip() or t.Fatal() (e.g., no RDMA devices in the CI environment), runtime.Goexit() runs only deferred functions. At that point only defer basens.Close() is registered, so:

• newns fd is leaked.
• The system RDMA mode remains "exclusive", potentially corrupting subsequent tests.

The 0.00s test duration in the CI failure is consistent with an early skip/bail at this point.

Move firstRdmaDevice(t) to the top of the test before any state change, or add explicit defers for cleanup before it is called:

🛠️ Proposed fix — hoist `firstRdmaDevice` before any state mutation

 func TestRdmaLinkSetNsFd(t *testing.T) {
 	minKernelRequired(t, 5, 2)
 	setupRdmaKModule(t, "ib_core")
+	link := firstRdmaDevice(t)
 
 	mode, err := RdmaSystemGetNetnsMode()
 	if err != nil {
 		t.Fatal(err)
 	}
 	t.Log("current rdma netns mode", mode)
 	err = RdmaSystemSetNetnsMode("exclusive")
 	if err != nil {
 		t.Fatal(err)
 	}
 	basens, err := netns.Get()
 	if err != nil {
 		RdmaSystemSetNetnsMode(mode)
 		t.Fatal("Failed to get basens")
 	}
 	defer basens.Close()
 
 	newns, err := netns.New()
 	if err != nil {
 		RdmaSystemSetNetnsMode(mode)
 		t.Fatal("Failed to create newns")
 	}
+	defer newns.Close()
 
 	netns.Set(basens)
-	link := firstRdmaDevice(t)
 	t.Log("rdma link: ", link)
 
 	err = RdmaLinkSetNsFd(link, uint32(newns))
 	if err != nil {
-		newns.Close()
 		RdmaSystemSetNetnsMode(mode)
 		t.Fatal(err)
 	}
 
-	newns.Close()
 	//Set the old mode back at start of the test
 	err = RdmaSystemSetNetnsMode(mode)
 	if err != nil {
 		t.Fatal(err)
 	}
 }

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@rdma_link_test.go` around lines 117 - 160, The test TestRdmaLinkSetNsFd
currently mutates global state (calls RdmaSystemSetNetnsMode("exclusive") and
creates newns) before calling firstRdmaDevice, which can t.Skip/t.Fatal and leak
newns and leave RDMA mode changed; move the call to firstRdmaDevice(t) to the
top of TestRdmaLinkSetNsFd before any calls to RdmaSystemSetNetnsMode or
netns.New, or if you must keep the order, immediately register cleanup defers
(defer newns.Close() and defer RdmaSystemSetNetnsMode(mode)) as soon as newns is
created and mode changed so that any early t.Skip/t.Fatal still runs the
cleanup.

rdma_link_test.go (1)

144-160: ⚠️ Potential issue | 🟠 Major

RdmaSystemSetNetnsMode(mode) cleanup skipped if firstRdmaDevice exits the test.

firstRdmaDevice(t) at line 144 may call t.Skip() (no devices) or t.Fatal() (list error). Both exit via runtime.Goexit(), which runs deferred calls but skips non-deferred code. Since the mode revert at line 156 is in the normal control flow and not deferred, the system is left in "exclusive" RDMA netns mode if firstRdmaDevice aborts the test at this point.

The fix is to defer the mode revert immediately after it is changed (line 126):

🛡️ Proposed fix

 	err = RdmaSystemSetNetnsMode("exclusive")
 	if err != nil {
 		t.Skipf("Failed to set RDMA netns mode to exclusive: %v", err)
 	}
+	defer RdmaSystemSetNetnsMode(mode)
 	basens, err := netns.Get()
 	if err != nil {
-		RdmaSystemSetNetnsMode(mode)
 		t.Fatal("Failed to get basens")
 	}
 	defer basens.Close()

 	newns, err := netns.New()
 	if err != nil {
-		RdmaSystemSetNetnsMode(mode)
 		t.Fatal("Failed to create newns")
 	}

 	netns.Set(basens)
 	link := firstRdmaDevice(t)
 	t.Log("rdma link: ", link)

 	err = RdmaLinkSetNsFd(link, uint32(newns))
 	if err != nil {
 		newns.Close()
-		RdmaSystemSetNetnsMode(mode)
 		t.Fatal(err)
 	}

 	newns.Close()
-	//Set the old mode back at start of the test
-	err = RdmaSystemSetNetnsMode(mode)
-	if err != nil {
-		t.Fatal(err)
-	}
 }

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@rdma_link_test.go` around lines 144 - 160, The test currently sets the RDMA
netns mode (via RdmaSystemSetNetnsMode) but later reverts it only in normal
control flow, so if firstRdmaDevice(t) calls t.Skip()/t.Fatal() the revert is
skipped; fix by deferring the revert immediately after changing the mode: right
after the call that switches mode (the earlier RdmaSystemSetNetnsMode call where
you capture the original mode in variable mode), add a defer that calls
RdmaSystemSetNetnsMode(mode) (handle/log any error inside the defer rather than
using t.Fatal), and similarly defer newns.Close() as soon as newns is created so
cleanup always runs even if firstRdmaDevice or subsequent calls like
RdmaLinkSetNsFd or RdmaLinkSetNsFd(link, uint32(newns)) cause the test to exit
early.