bpf: add tail call tests to test_verifier

One of the downsides of the test_bpf module was that since being
in kernel space, it couldn't test-run tail calls. Now that the
test_verifier has the ability to perform run-time tests, populate
the prog array so we actually jump into other BPF programs and
can check all corner cases. Most useful in combination with JITs.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

tools/testing/selftests/bpf/test_verifier.c

@@ -2588,6 +2588,62 @@ static struct bpf_test tests[] = {
 		.result_unpriv = REJECT,
 		.result = ACCEPT,
 	},
+	{
+		"runtime/jit: tail_call within bounds, prog once",
+		.insns = {
+			BPF_MOV64_IMM(BPF_REG_3, 0),
+			BPF_LD_MAP_FD(BPF_REG_2, 0),
+			BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
+				     BPF_FUNC_tail_call),
+			BPF_MOV64_IMM(BPF_REG_0, 1),
+			BPF_EXIT_INSN(),
+		},
+		.fixup_prog = { 1 },
+		.result = ACCEPT,
+		.retval = 42,
+	},
+	{
+		"runtime/jit: tail_call within bounds, prog loop",
+		.insns = {
+			BPF_MOV64_IMM(BPF_REG_3, 1),
+			BPF_LD_MAP_FD(BPF_REG_2, 0),
+			BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
+				     BPF_FUNC_tail_call),
+			BPF_MOV64_IMM(BPF_REG_0, 1),
+			BPF_EXIT_INSN(),
+		},
+		.fixup_prog = { 1 },
+		.result = ACCEPT,
+		.retval = 41,
+	},
+	{
+		"runtime/jit: tail_call within bounds, no prog",
+		.insns = {
+			BPF_MOV64_IMM(BPF_REG_3, 2),
+			BPF_LD_MAP_FD(BPF_REG_2, 0),
+			BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
+				     BPF_FUNC_tail_call),
+			BPF_MOV64_IMM(BPF_REG_0, 1),
+			BPF_EXIT_INSN(),
+		},
+		.fixup_prog = { 1 },
+		.result = ACCEPT,
+		.retval = 1,
+	},
+	{
+		"runtime/jit: tail_call out of bounds",
+		.insns = {
+			BPF_MOV64_IMM(BPF_REG_3, 256),
+			BPF_LD_MAP_FD(BPF_REG_2, 0),
+			BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
+				     BPF_FUNC_tail_call),
+			BPF_MOV64_IMM(BPF_REG_0, 2),
+			BPF_EXIT_INSN(),
+		},
+		.fixup_prog = { 1 },
+		.result = ACCEPT,
+		.retval = 2,
+	},
 	{
 		"runtime/jit: pass negative index to tail_call",
 		.insns = {
@@ -2595,11 +2651,12 @@ static struct bpf_test tests[] = {
 			BPF_LD_MAP_FD(BPF_REG_2, 0),
 			BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
 				     BPF_FUNC_tail_call),
-			BPF_MOV64_IMM(BPF_REG_0, 0),
+			BPF_MOV64_IMM(BPF_REG_0, 2),
 			BPF_EXIT_INSN(),
 		},
 		.fixup_prog = { 1 },
 		.result = ACCEPT,
+		.retval = 2,
 	},
 	{
 		"runtime/jit: pass > 32bit index to tail_call",
@@ -2608,11 +2665,12 @@ static struct bpf_test tests[] = {
 			BPF_LD_MAP_FD(BPF_REG_2, 0),
 			BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
 				     BPF_FUNC_tail_call),
-			BPF_MOV64_IMM(BPF_REG_0, 0),
+			BPF_MOV64_IMM(BPF_REG_0, 2),
 			BPF_EXIT_INSN(),
 		},
 		.fixup_prog = { 2 },
 		.result = ACCEPT,
+		.retval = 42,
 	},
 	{
 		"stack pointer arithmetic",
@@ -11278,16 +11336,61 @@ static int create_map(uint32_t size_value, uint32_t max_elem)
 	return fd;
 }
 
+static int create_prog_dummy1(void)
+{
+	struct bpf_insn prog[] = {
+		BPF_MOV64_IMM(BPF_REG_0, 42),
+		BPF_EXIT_INSN(),
+	};
+
+	return bpf_load_program(BPF_PROG_TYPE_SOCKET_FILTER, prog,
+				ARRAY_SIZE(prog), "GPL", 0, NULL, 0);
+}
+
+static int create_prog_dummy2(int mfd, int idx)
+{
+	struct bpf_insn prog[] = {
+		BPF_MOV64_IMM(BPF_REG_3, idx),
+		BPF_LD_MAP_FD(BPF_REG_2, mfd),
+		BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
+			     BPF_FUNC_tail_call),
+		BPF_MOV64_IMM(BPF_REG_0, 41),
+		BPF_EXIT_INSN(),
+	};
+
+	return bpf_load_program(BPF_PROG_TYPE_SOCKET_FILTER, prog,
+				ARRAY_SIZE(prog), "GPL", 0, NULL, 0);
+}
+
 static int create_prog_array(void)
 {
-	int fd;
+	int p1key = 0, p2key = 1;
+	int mfd, p1fd, p2fd;
 
-	fd = bpf_create_map(BPF_MAP_TYPE_PROG_ARRAY, sizeof(int),
-			    sizeof(int), 4, 0);
-	if (fd < 0)
+	mfd = bpf_create_map(BPF_MAP_TYPE_PROG_ARRAY, sizeof(int),
+			     sizeof(int), 4, 0);
+	if (mfd < 0) {
 		printf("Failed to create prog array '%s'!\n", strerror(errno));
+		return -1;
+	}
 
-	return fd;
+	p1fd = create_prog_dummy1();
+	p2fd = create_prog_dummy2(mfd, p2key);
+	if (p1fd < 0 || p2fd < 0)
+		goto out;
+	if (bpf_map_update_elem(mfd, &p1key, &p1fd, BPF_ANY) < 0)
+		goto out;
+	if (bpf_map_update_elem(mfd, &p2key, &p2fd, BPF_ANY) < 0)
+		goto out;
+	close(p2fd);
+	close(p1fd);
+
+	return mfd;
+out:
+	close(p2fd);
+	close(p1fd);
+	close(mfd);
+	return -1;
 }
 
 static int create_map_in_map(void)