test(pipe): lock in pipe/pipe2 IPC classification and fd-pair exit reads

Audit of pipe(2)/pipe2(2) (task dx) confirmed the tracing implementation is
correct: KindPipe (not KindFd, since args[0] is an output ptr to int[2], not an
fd), FamilyIPC, and an UNCLASSIFIED int return. Enter stashes the output ptr
(flags=0 for pipe, args[1] for pipe2); exit reads the fd pair via
bpf_probe_read_user guarded by ret==0, mirroring the socketpair pipe-like
pattern. The only gaps were missing lock-in tests, now added:

- codegen: assert the exit handler reads the fd pair from the stashed output
  buffer (ret==0 guard, bpf_probe_read_user, fd0/fd1) and that the flag-less
  pipe variant hardcodes flags=0 and never reads args[1].
- classify: pipe/pipe2 are never KindFd and stay UNCLASSIFIED on ret.
- runtime: a failed pipe (ret==-1) tracks no descriptors and attaches no file.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

2 files changed, 70 insertions, 0 deletions

diff --git a/internal/generate/classify_test.go b/internal/generate/classify_test.go
index 3aea4fe..514c2e8 100644
--- a/internal/generate/classify_test.go
+++ b/internal/generate/classify_test.go
@@ -890,6 +890,41 @@ func TestClassifyExitPipe2(t *testing.T) {
 	}
 }
 
+// TestClassifyPipeNotFd locks in that pipe(2) is NOT classified as KindFd.
+// pipe's args[0] is an OUTPUT pointer to int[2] (the two created fds are written
+// there by the kernel and are only valid AFTER the syscall returns), NOT an fd
+// argument. Capturing args[0] as an fd would attribute the pipe to a bogus
+// descriptor; pipe must use the pipe-specific KindPipe path that reads the fd
+// pair from the userspace buffer at exit. Same pitfall as socketpair (task c00).
+func TestClassifyPipeNotFd(t *testing.T) {
+	for _, name := range []string{"pipe", "pipe2"} {
+		r := classifyFromData(t, map[string]string{
+			"pipe":  FormatPipe,
+			"pipe2": FormatPipe2,
+		}[name])
+		if r.Kind == KindFd {
+			t.Fatalf("%s classified as KindFd: args[0] is an output ptr, not an fd", name)
+		}
+		if r.Kind != KindPipe {
+			t.Errorf("%s: got kind %d, want KindPipe", name, r.Kind)
+		}
+	}
+}
+
+// TestClassifyPipeUnclassifiedRet locks in that the pipe and pipe2 exit
+// tracepoints stay UNCLASSIFIED. pipe(2)/pipe2(2) return int (0 on success,
+// -1 on error) — a status code, NOT a transferred byte count. They must not be
+// in retClassifications and must never map to READ/WRITE/TRANSFER, which would
+// misreport phantom bytes. The created fds are surfaced via fd0/fd1 in the
+// pipe_event, not via the return value.
+func TestClassifyPipeUnclassifiedRet(t *testing.T) {
+	for _, name := range []string{"sys_exit_pipe", "sys_exit_pipe2"} {
+		if got := ClassifyRet(name); got != Unclassified {
+			t.Errorf("ClassifyRet(%s) = %q, want UNCLASSIFIED", name, got)
+		}
+	}
+}
+
 func TestClassifyEventfd(t *testing.T) {
 	r := classifyFromData(t, FormatEventfd)
 	if r.Kind != KindEventfd {
diff --git a/internal/generate/codegen_test.go b/internal/generate/codegen_test.go
index 58ed60c..9545447 100644
--- a/internal/generate/codegen_test.go
+++ b/internal/generate/codegen_test.go
@@ -1650,6 +1650,41 @@ func TestGeneratePipeHandler(t *testing.T) {
 	requireContains(t, output, "ev->ret = ctx->ret;")
 }
 
+// TestGeneratePipeHandlerExitReadsFdPair locks in the pipe-specific exit path:
+// args[0] is an OUTPUT pointer to int[2], not an fd. The two created fds are
+// only valid AFTER the syscall returns, so the exit handler must read them from
+// the stashed userspace buffer with bpf_probe_read_user, guarded by ret == 0
+// (a failed pipe(2) leaves the buffer untouched). This mirrors the socketpair
+// audit (task c00) pipe-like pattern.
+func TestGeneratePipeHandlerExitReadsFdPair(t *testing.T) {
+	output := generateFromPair(t, FormatPipe2, FormatExitPipe2)
+
+	// Exit reads the fd pair from the stashed output pointer only on success.
+	requireContains(t, output, "if (ctx->ret == 0 && pending->upipefd != 0) {")
+	requireContains(t, output, "int pipefd[2];")
+	requireContains(t, output, "bpf_probe_read_user(&pipefd, sizeof(pipefd), (void *)pending->upipefd)")
+	requireContains(t, output, "fd0 = (__s32)pipefd[0];")
+	requireContains(t, output, "fd1 = (__s32)pipefd[1];")
+	requireContains(t, output, "bpf_map_delete_elem(&pipe_ctx_map, &tid);")
+	requireContains(t, output, "ev->fd0 = fd0;")
+	requireContains(t, output, "ev->fd1 = fd1;")
+}
+
+// TestGeneratePlainPipeHandlerZeroFlags locks in that the flag-less pipe(2)
+// variant hardcodes flags = 0 and never reads args[1] (which does not exist for
+// pipe; only pipe2 has a flags argument at args[1]).
+func TestGeneratePlainPipeHandlerZeroFlags(t *testing.T) {
+	output := generateFromPair(t, FormatPipe, FormatExitPipe)
+
+	requireContains(t, output, "struct pipe_event *ev")
+	requireContains(t, output, "ev->event_type = ENTER_PIPE_EVENT;")
+	requireContains(t, output, "pending.upipefd = ctx->args[0];")
+	requireContains(t, output, "pending.flags = 0;")
+	requireNotContains(t, output, "pending.flags = (__s32)ctx->args[1];")
+	requireContains(t, output, "SEC(\"tracepoint/syscalls/sys_exit_pipe\")")
+	requireContains(t, output, "ev->event_type = EXIT_PIPE_EVENT;")
+}
+
 func TestGenerateEventfdHandler(t *testing.T) {
 	output := generateFromPair(t, FormatEventfd2, FormatExitEventfd2)