]> git.baikalelectronics.ru Git - kernel.git/commitdiff
selftests/bpf: Extend test_flow_dissector to cover link creation
authorJakub Sitnicki <jakub@cloudflare.com>
Sun, 31 May 2020 08:28:46 +0000 (10:28 +0200)
committerAlexei Starovoitov <ast@kernel.org>
Mon, 1 Jun 2020 22:21:03 +0000 (15:21 -0700)
Extend the existing flow_dissector test case to run tests once using direct
prog attachments, and then for the second time using indirect attachment
via link.

The intention is to exercises the newly added high-level API for attaching
programs to network namespace with links (bpf_program__attach_netns).

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200531082846.2117903-13-jakub@cloudflare.com
tools/testing/selftests/bpf/prog_tests/flow_dissector.c

index b6370c0b3b7af776946c408f02e0b6616ab4b041..ea14e3ece81202ea8e4d922a1e0b041b2f2e2c99 100644 (file)
@@ -103,6 +103,7 @@ struct test {
 
 #define VLAN_HLEN      4
 
+static __u32 duration;
 struct test tests[] = {
        {
                .name = "ipv4",
@@ -474,11 +475,87 @@ static int init_prog_array(struct bpf_object *obj, struct bpf_map *prog_array)
        return 0;
 }
 
+static void run_tests_skb_less(int tap_fd, struct bpf_map *keys)
+{
+       int i, err, keys_fd;
+
+       keys_fd = bpf_map__fd(keys);
+       if (CHECK(keys_fd < 0, "bpf_map__fd", "err %d\n", keys_fd))
+               return;
+
+       for (i = 0; i < ARRAY_SIZE(tests); i++) {
+               /* Keep in sync with 'flags' from eth_get_headlen. */
+               __u32 eth_get_headlen_flags =
+                       BPF_FLOW_DISSECTOR_F_PARSE_1ST_FRAG;
+               struct bpf_prog_test_run_attr tattr = {};
+               struct bpf_flow_keys flow_keys = {};
+               __u32 key = (__u32)(tests[i].keys.sport) << 16 |
+                           tests[i].keys.dport;
+
+               /* For skb-less case we can't pass input flags; run
+                * only the tests that have a matching set of flags.
+                */
+
+               if (tests[i].flags != eth_get_headlen_flags)
+                       continue;
+
+               err = tx_tap(tap_fd, &tests[i].pkt, sizeof(tests[i].pkt));
+               CHECK(err < 0, "tx_tap", "err %d errno %d\n", err, errno);
+
+               err = bpf_map_lookup_elem(keys_fd, &key, &flow_keys);
+               CHECK_ATTR(err, tests[i].name, "bpf_map_lookup_elem %d\n", err);
+
+               CHECK_ATTR(err, tests[i].name, "skb-less err %d\n", err);
+               CHECK_FLOW_KEYS(tests[i].name, flow_keys, tests[i].keys);
+
+               err = bpf_map_delete_elem(keys_fd, &key);
+               CHECK_ATTR(err, tests[i].name, "bpf_map_delete_elem %d\n", err);
+       }
+}
+
+static void test_skb_less_prog_attach(struct bpf_flow *skel, int tap_fd)
+{
+       int err, prog_fd;
+
+       prog_fd = bpf_program__fd(skel->progs._dissect);
+       if (CHECK(prog_fd < 0, "bpf_program__fd", "err %d\n", prog_fd))
+               return;
+
+       err = bpf_prog_attach(prog_fd, 0, BPF_FLOW_DISSECTOR, 0);
+       if (CHECK(err, "bpf_prog_attach", "err %d errno %d\n", err, errno))
+               return;
+
+       run_tests_skb_less(tap_fd, skel->maps.last_dissection);
+
+       err = bpf_prog_detach(prog_fd, BPF_FLOW_DISSECTOR);
+       CHECK(err, "bpf_prog_detach", "err %d errno %d\n", err, errno);
+}
+
+static void test_skb_less_link_create(struct bpf_flow *skel, int tap_fd)
+{
+       struct bpf_link *link;
+       int err, net_fd;
+
+       net_fd = open("/proc/self/ns/net", O_RDONLY);
+       if (CHECK(net_fd < 0, "open(/proc/self/ns/net)", "err %d\n", errno))
+               return;
+
+       link = bpf_program__attach_netns(skel->progs._dissect, net_fd);
+       if (CHECK(IS_ERR(link), "attach_netns", "err %ld\n", PTR_ERR(link)))
+               goto out_close;
+
+       run_tests_skb_less(tap_fd, skel->maps.last_dissection);
+
+       err = bpf_link__destroy(link);
+       CHECK(err, "bpf_link__destroy", "err %d\n", err);
+out_close:
+       close(net_fd);
+}
+
 void test_flow_dissector(void)
 {
        int i, err, prog_fd, keys_fd = -1, tap_fd;
        struct bpf_flow *skel;
-       __u32 duration = 0;
 
        skel = bpf_flow__open_and_load();
        if (CHECK(!skel, "skel", "failed to open/load skeleton\n"))
@@ -526,45 +603,17 @@ void test_flow_dissector(void)
         * via BPF map in this case.
         */
 
-       err = bpf_prog_attach(prog_fd, 0, BPF_FLOW_DISSECTOR, 0);
-       CHECK(err, "bpf_prog_attach", "err %d errno %d\n", err, errno);
-
        tap_fd = create_tap("tap0");
        CHECK(tap_fd < 0, "create_tap", "tap_fd %d errno %d\n", tap_fd, errno);
        err = ifup("tap0");
        CHECK(err, "ifup", "err %d errno %d\n", err, errno);
 
-       for (i = 0; i < ARRAY_SIZE(tests); i++) {
-               /* Keep in sync with 'flags' from eth_get_headlen. */
-               __u32 eth_get_headlen_flags =
-                       BPF_FLOW_DISSECTOR_F_PARSE_1ST_FRAG;
-               struct bpf_prog_test_run_attr tattr = {};
-               struct bpf_flow_keys flow_keys = {};
-               __u32 key = (__u32)(tests[i].keys.sport) << 16 |
-                           tests[i].keys.dport;
-
-               /* For skb-less case we can't pass input flags; run
-                * only the tests that have a matching set of flags.
-                */
-
-               if (tests[i].flags != eth_get_headlen_flags)
-                       continue;
-
-               err = tx_tap(tap_fd, &tests[i].pkt, sizeof(tests[i].pkt));
-               CHECK(err < 0, "tx_tap", "err %d errno %d\n", err, errno);
-
-               err = bpf_map_lookup_elem(keys_fd, &key, &flow_keys);
-               CHECK_ATTR(err, tests[i].name, "bpf_map_lookup_elem %d\n", err);
-
-               CHECK_ATTR(err, tests[i].name, "skb-less err %d\n", err);
-               CHECK_FLOW_KEYS(tests[i].name, flow_keys, tests[i].keys);
-
-               err = bpf_map_delete_elem(keys_fd, &key);
-               CHECK_ATTR(err, tests[i].name, "bpf_map_delete_elem %d\n", err);
-       }
+       /* Test direct prog attachment */
+       test_skb_less_prog_attach(skel, tap_fd);
+       /* Test indirect prog attachment via link */
+       test_skb_less_link_create(skel, tap_fd);
 
        close(tap_fd);
-       bpf_prog_detach(prog_fd, BPF_FLOW_DISSECTOR);
 out_destroy_skel:
        bpf_flow__destroy(skel);
 }