Quelle  tc_links.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0
/* Copyright (c) 2023 Isovalent */
#include <uapi/linux/if_link.h>
#include <uapi/linux/pkt_sched.h>
#include <net/if.h>
#include <test_progs.h>

#define loopback 1
#define ping_cmd "ping -q -c1 -w1 127.0.0.1 > /dev/null"

#include "test_tc_link.skel.h"

#include "netlink_helpers.h"
#include "tc_helpers.h"

void test_ns_tc_links_basic(void)
{
 LIBBPF_OPTS(bpf_prog_query_opts, optq);
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 prog_ids[2], link_ids[2];
 __u32 pid1, pid2, lid1, lid2;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err;

 skel = test_tc_link__open_and_load();
 if (!ASSERT_OK_PTR(skel, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");

 assert_mprog_count(BPF_TCX_INGRESS, 0);
 assert_mprog_count(BPF_TCX_EGRESS, 0);

 ASSERT_EQ(skel->bss->seen_tc1, false, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, false, "seen_tc2");

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 lid1 = id_from_link_fd(bpf_link__fd(skel->links.tc1));

 assert_mprog_count(BPF_TCX_INGRESS, 1);
 assert_mprog_count(BPF_TCX_EGRESS, 0);

 optq.prog_ids = prog_ids;
 optq.link_ids = link_ids;

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, BPF_TCX_INGRESS, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 1, "count");
 ASSERT_EQ(optq.revision, 2, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid1, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid1, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], 0, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], 0, "link_ids[1]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, false, "seen_tc2");

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 lid2 = id_from_link_fd(bpf_link__fd(skel->links.tc2));
 ASSERT_NEQ(lid1, lid2, "link_ids_1_2");

 assert_mprog_count(BPF_TCX_INGRESS, 1);
 assert_mprog_count(BPF_TCX_EGRESS, 1);

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, BPF_TCX_EGRESS, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 1, "count");
 ASSERT_EQ(optq.revision, 2, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid2, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid2, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], 0, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], 0, "link_ids[1]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
cleanup:
 test_tc_link__destroy(skel);

 assert_mprog_count(BPF_TCX_INGRESS, 0);
 assert_mprog_count(BPF_TCX_EGRESS, 0);
}

static void test_tc_links_before_target(int target)
{
 LIBBPF_OPTS(bpf_prog_query_opts, optq);
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 prog_ids[5], link_ids[5];
 __u32 pid1, pid2, pid3, pid4;
 __u32 lid1, lid2, lid3, lid4;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err;

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc3, target),
    0, "tc3_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc4, target),
    0, "tc4_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));
 pid3 = id_from_prog_fd(bpf_program__fd(skel->progs.tc3));
 pid4 = id_from_prog_fd(bpf_program__fd(skel->progs.tc4));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");
 ASSERT_NEQ(pid3, pid4, "prog_ids_3_4");
 ASSERT_NEQ(pid2, pid3, "prog_ids_2_3");

 assert_mprog_count(target, 0);

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 lid1 = id_from_link_fd(bpf_link__fd(skel->links.tc1));

 assert_mprog_count(target, 1);

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 lid2 = id_from_link_fd(bpf_link__fd(skel->links.tc2));

 assert_mprog_count(target, 2);

 optq.prog_ids = prog_ids;
 optq.link_ids = link_ids;

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 2, "count");
 ASSERT_EQ(optq.revision, 3, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid1, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid1, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid2, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid2, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], 0, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], 0, "link_ids[2]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, false, "seen_tc3");
 ASSERT_EQ(skel->bss->seen_tc4, false, "seen_tc4");

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE,
  .relative_fd = bpf_program__fd(skel->progs.tc2),
 );

 link = bpf_program__attach_tcx(skel->progs.tc3, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc3 = link;

 lid3 = id_from_link_fd(bpf_link__fd(skel->links.tc3));

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE | BPF_F_LINK,
  .relative_id = lid1,
 );

 link = bpf_program__attach_tcx(skel->progs.tc4, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc4 = link;

 lid4 = id_from_link_fd(bpf_link__fd(skel->links.tc4));

 assert_mprog_count(target, 4);

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 4, "count");
 ASSERT_EQ(optq.revision, 5, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid4, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid4, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid1, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid1, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], pid3, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], lid3, "link_ids[2]");
 ASSERT_EQ(optq.prog_ids[3], pid2, "prog_ids[3]");
 ASSERT_EQ(optq.link_ids[3], lid2, "link_ids[3]");
 ASSERT_EQ(optq.prog_ids[4], 0, "prog_ids[4]");
 ASSERT_EQ(optq.link_ids[4], 0, "link_ids[4]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, true, "seen_tc3");
 ASSERT_EQ(skel->bss->seen_tc4, true, "seen_tc4");
cleanup:
 test_tc_link__destroy(skel);
 assert_mprog_count(target, 0);
}

void test_ns_tc_links_before(void)
{
 test_tc_links_before_target(BPF_TCX_INGRESS);
 test_tc_links_before_target(BPF_TCX_EGRESS);
}

static void test_tc_links_after_target(int target)
{
 LIBBPF_OPTS(bpf_prog_query_opts, optq);
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 prog_ids[5], link_ids[5];
 __u32 pid1, pid2, pid3, pid4;
 __u32 lid1, lid2, lid3, lid4;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err;

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc3, target),
    0, "tc3_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc4, target),
    0, "tc4_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));
 pid3 = id_from_prog_fd(bpf_program__fd(skel->progs.tc3));
 pid4 = id_from_prog_fd(bpf_program__fd(skel->progs.tc4));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");
 ASSERT_NEQ(pid3, pid4, "prog_ids_3_4");
 ASSERT_NEQ(pid2, pid3, "prog_ids_2_3");

 assert_mprog_count(target, 0);

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 lid1 = id_from_link_fd(bpf_link__fd(skel->links.tc1));

 assert_mprog_count(target, 1);

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 lid2 = id_from_link_fd(bpf_link__fd(skel->links.tc2));

 assert_mprog_count(target, 2);

 optq.prog_ids = prog_ids;
 optq.link_ids = link_ids;

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 2, "count");
 ASSERT_EQ(optq.revision, 3, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid1, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid1, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid2, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid2, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], 0, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], 0, "link_ids[2]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, false, "seen_tc3");
 ASSERT_EQ(skel->bss->seen_tc4, false, "seen_tc4");

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_AFTER,
  .relative_fd = bpf_program__fd(skel->progs.tc1),
 );

 link = bpf_program__attach_tcx(skel->progs.tc3, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc3 = link;

 lid3 = id_from_link_fd(bpf_link__fd(skel->links.tc3));

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_AFTER | BPF_F_LINK,
  .relative_fd = bpf_link__fd(skel->links.tc2),
 );

 link = bpf_program__attach_tcx(skel->progs.tc4, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc4 = link;

 lid4 = id_from_link_fd(bpf_link__fd(skel->links.tc4));

 assert_mprog_count(target, 4);

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 4, "count");
 ASSERT_EQ(optq.revision, 5, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid1, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid1, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid3, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid3, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], pid2, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], lid2, "link_ids[2]");
 ASSERT_EQ(optq.prog_ids[3], pid4, "prog_ids[3]");
 ASSERT_EQ(optq.link_ids[3], lid4, "link_ids[3]");
 ASSERT_EQ(optq.prog_ids[4], 0, "prog_ids[4]");
 ASSERT_EQ(optq.link_ids[4], 0, "link_ids[4]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, true, "seen_tc3");
 ASSERT_EQ(skel->bss->seen_tc4, true, "seen_tc4");
cleanup:
 test_tc_link__destroy(skel);
 assert_mprog_count(target, 0);
}

void test_ns_tc_links_after(void)
{
 test_tc_links_after_target(BPF_TCX_INGRESS);
 test_tc_links_after_target(BPF_TCX_EGRESS);
}

static void test_tc_links_revision_target(int target)
{
 LIBBPF_OPTS(bpf_prog_query_opts, optq);
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 prog_ids[3], link_ids[3];
 __u32 pid1, pid2, lid1, lid2;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err;

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");

 assert_mprog_count(target, 0);

 optl.expected_revision = 1;

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 lid1 = id_from_link_fd(bpf_link__fd(skel->links.tc1));

 assert_mprog_count(target, 1);

 optl.expected_revision = 1;

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 1);

 optl.expected_revision = 2;

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 lid2 = id_from_link_fd(bpf_link__fd(skel->links.tc2));

 assert_mprog_count(target, 2);

 optq.prog_ids = prog_ids;
 optq.link_ids = link_ids;

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 2, "count");
 ASSERT_EQ(optq.revision, 3, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid1, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid1, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid2, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid2, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], 0, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], 0, "prog_ids[2]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
cleanup:
 test_tc_link__destroy(skel);
 assert_mprog_count(target, 0);
}

void test_ns_tc_links_revision(void)
{
 test_tc_links_revision_target(BPF_TCX_INGRESS);
 test_tc_links_revision_target(BPF_TCX_EGRESS);
}

static void test_tc_chain_classic(int target, bool chain_tc_old)
{
 LIBBPF_OPTS(bpf_tc_opts, tc_opts, .handle = 1, .priority = 1);
 LIBBPF_OPTS(bpf_tc_hook, tc_hook, .ifindex = loopback);
 bool hook_created = false, tc_attached = false;
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 pid1, pid2, pid3;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err;

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));
 pid3 = id_from_prog_fd(bpf_program__fd(skel->progs.tc3));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");
 ASSERT_NEQ(pid2, pid3, "prog_ids_2_3");

 assert_mprog_count(target, 0);

 if (chain_tc_old) {
  tc_hook.attach_point = target == BPF_TCX_INGRESS ?
           BPF_TC_INGRESS : BPF_TC_EGRESS;
  err = bpf_tc_hook_create(&tc_hook);
  if (err == 0)
   hook_created = true;
  err = err == -EEXIST ? 0 : err;
  if (!ASSERT_OK(err, "bpf_tc_hook_create"))
   goto cleanup;

  tc_opts.prog_fd = bpf_program__fd(skel->progs.tc3);
  err = bpf_tc_attach(&tc_hook, &tc_opts);
  if (!ASSERT_OK(err, "bpf_tc_attach"))
   goto cleanup;
  tc_attached = true;
 }

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 assert_mprog_count(target, 2);

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, chain_tc_old, "seen_tc3");

 err = bpf_link__detach(skel->links.tc2);
 if (!ASSERT_OK(err, "prog_detach"))
  goto cleanup;

 assert_mprog_count(target, 1);

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, false, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, chain_tc_old, "seen_tc3");
cleanup:
 if (tc_attached) {
  tc_opts.flags = tc_opts.prog_fd = tc_opts.prog_id = 0;
  err = bpf_tc_detach(&tc_hook, &tc_opts);
  ASSERT_OK(err, "bpf_tc_detach");
 }
 if (hook_created) {
  tc_hook.attach_point = BPF_TC_INGRESS | BPF_TC_EGRESS;
  bpf_tc_hook_destroy(&tc_hook);
 }
 assert_mprog_count(target, 1);
 test_tc_link__destroy(skel);
 assert_mprog_count(target, 0);
}

void test_ns_tc_links_chain_classic(void)
{
 test_tc_chain_classic(BPF_TCX_INGRESS, false);
 test_tc_chain_classic(BPF_TCX_EGRESS, false);
 test_tc_chain_classic(BPF_TCX_INGRESS, true);
 test_tc_chain_classic(BPF_TCX_EGRESS, true);
}

static void test_tc_links_replace_target(int target)
{
 LIBBPF_OPTS(bpf_prog_query_opts, optq);
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 pid1, pid2, pid3, lid1, lid2;
 __u32 prog_ids[4], link_ids[4];
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err;

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc3, target),
    0, "tc3_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));
 pid3 = id_from_prog_fd(bpf_program__fd(skel->progs.tc3));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");
 ASSERT_NEQ(pid2, pid3, "prog_ids_2_3");

 assert_mprog_count(target, 0);

 optl.expected_revision = 1;

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 lid1 = id_from_link_fd(bpf_link__fd(skel->links.tc1));

 assert_mprog_count(target, 1);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE,
  .relative_id = pid1,
  .expected_revision = 2,
 );

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 lid2 = id_from_link_fd(bpf_link__fd(skel->links.tc2));

 assert_mprog_count(target, 2);

 optq.prog_ids = prog_ids;
 optq.link_ids = link_ids;

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 2, "count");
 ASSERT_EQ(optq.revision, 3, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid2, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid2, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid1, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid1, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], 0, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], 0, "link_ids[2]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, false, "seen_tc3");

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_REPLACE,
  .relative_fd = bpf_program__fd(skel->progs.tc2),
  .expected_revision = 3,
 );

 link = bpf_program__attach_tcx(skel->progs.tc3, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 2);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_REPLACE | BPF_F_LINK,
  .relative_fd = bpf_link__fd(skel->links.tc2),
  .expected_revision = 3,
 );

 link = bpf_program__attach_tcx(skel->progs.tc3, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 2);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_REPLACE | BPF_F_LINK | BPF_F_AFTER,
  .relative_id = lid2,
 );

 link = bpf_program__attach_tcx(skel->progs.tc3, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 2);

 err = bpf_link__update_program(skel->links.tc2, skel->progs.tc3);
 if (!ASSERT_OK(err, "link_update"))
  goto cleanup;

 assert_mprog_count(target, 2);

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 2, "count");
 ASSERT_EQ(optq.revision, 4, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid3, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid2, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid1, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid1, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], 0, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], 0, "link_ids[2]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, false, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, true, "seen_tc3");

 err = bpf_link__detach(skel->links.tc2);
 if (!ASSERT_OK(err, "link_detach"))
  goto cleanup;

 assert_mprog_count(target, 1);

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 1, "count");
 ASSERT_EQ(optq.revision, 5, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid1, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid1, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], 0, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], 0, "link_ids[1]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, false, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, false, "seen_tc3");

 err = bpf_link__update_program(skel->links.tc1, skel->progs.tc1);
 if (!ASSERT_OK(err, "link_update_self"))
  goto cleanup;

 assert_mprog_count(target, 1);

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 1, "count");
 ASSERT_EQ(optq.revision, 5, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid1, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid1, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], 0, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], 0, "link_ids[1]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, false, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, false, "seen_tc3");
cleanup:
 test_tc_link__destroy(skel);
 assert_mprog_count(target, 0);
}

void test_ns_tc_links_replace(void)
{
 test_tc_links_replace_target(BPF_TCX_INGRESS);
 test_tc_links_replace_target(BPF_TCX_EGRESS);
}

static void test_tc_links_invalid_target(int target)
{
 LIBBPF_OPTS(bpf_prog_query_opts, optq);
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 pid1, pid2, lid1;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err;

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");

 assert_mprog_count(target, 0);

 optl.flags = BPF_F_BEFORE | BPF_F_AFTER;

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE | BPF_F_ID,
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_AFTER | BPF_F_ID,
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_ID,
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_LINK,
  .relative_fd = bpf_program__fd(skel->progs.tc2),
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_LINK,
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .relative_fd = bpf_program__fd(skel->progs.tc2),
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE | BPF_F_AFTER,
  .relative_fd = bpf_program__fd(skel->progs.tc2),
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE,
  .relative_fd = bpf_program__fd(skel->progs.tc1),
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_ID,
  .relative_id = pid2,
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_ID,
  .relative_id = 42,
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE,
  .relative_fd = bpf_program__fd(skel->progs.tc1),
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE | BPF_F_LINK,
  .relative_fd = bpf_program__fd(skel->progs.tc1),
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_AFTER,
  .relative_fd = bpf_program__fd(skel->progs.tc1),
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl);

 link = bpf_program__attach_tcx(skel->progs.tc1, 0, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_AFTER | BPF_F_LINK,
  .relative_fd = bpf_program__fd(skel->progs.tc1),
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 0);

 LIBBPF_OPTS_RESET(optl);

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 lid1 = id_from_link_fd(bpf_link__fd(skel->links.tc1));

 assert_mprog_count(target, 1);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_AFTER | BPF_F_LINK,
  .relative_fd = bpf_program__fd(skel->progs.tc1),
 );

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 1);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE | BPF_F_LINK | BPF_F_ID,
  .relative_id = ~0,
 );

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 1);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE | BPF_F_LINK | BPF_F_ID,
  .relative_id = lid1,
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }

 assert_mprog_count(target, 1);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE | BPF_F_ID,
  .relative_id = pid1,
 );

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_ERR_PTR(link, "link_attach_should_fail")) {
  bpf_link__destroy(link);
  goto cleanup;
 }
 assert_mprog_count(target, 1);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE | BPF_F_LINK | BPF_F_ID,
  .relative_id = lid1,
 );

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 assert_mprog_count(target, 2);
cleanup:
 test_tc_link__destroy(skel);
 assert_mprog_count(target, 0);
}

void test_ns_tc_links_invalid(void)
{
 test_tc_links_invalid_target(BPF_TCX_INGRESS);
 test_tc_links_invalid_target(BPF_TCX_EGRESS);
}

static void test_tc_links_prepend_target(int target)
{
 LIBBPF_OPTS(bpf_prog_query_opts, optq);
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 prog_ids[5], link_ids[5];
 __u32 pid1, pid2, pid3, pid4;
 __u32 lid1, lid2, lid3, lid4;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err;

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc3, target),
    0, "tc3_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc4, target),
    0, "tc4_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));
 pid3 = id_from_prog_fd(bpf_program__fd(skel->progs.tc3));
 pid4 = id_from_prog_fd(bpf_program__fd(skel->progs.tc4));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");
 ASSERT_NEQ(pid3, pid4, "prog_ids_3_4");
 ASSERT_NEQ(pid2, pid3, "prog_ids_2_3");

 assert_mprog_count(target, 0);

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 lid1 = id_from_link_fd(bpf_link__fd(skel->links.tc1));

 assert_mprog_count(target, 1);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE,
 );

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 lid2 = id_from_link_fd(bpf_link__fd(skel->links.tc2));

 assert_mprog_count(target, 2);

 optq.prog_ids = prog_ids;
 optq.link_ids = link_ids;

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 2, "count");
 ASSERT_EQ(optq.revision, 3, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid2, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid2, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid1, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid1, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], 0, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], 0, "link_ids[2]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, false, "seen_tc3");
 ASSERT_EQ(skel->bss->seen_tc4, false, "seen_tc4");

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE,
 );

 link = bpf_program__attach_tcx(skel->progs.tc3, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc3 = link;

 lid3 = id_from_link_fd(bpf_link__fd(skel->links.tc3));

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_BEFORE,
 );

 link = bpf_program__attach_tcx(skel->progs.tc4, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc4 = link;

 lid4 = id_from_link_fd(bpf_link__fd(skel->links.tc4));

 assert_mprog_count(target, 4);

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 4, "count");
 ASSERT_EQ(optq.revision, 5, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid4, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid4, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid3, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid3, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], pid2, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], lid2, "link_ids[2]");
 ASSERT_EQ(optq.prog_ids[3], pid1, "prog_ids[3]");
 ASSERT_EQ(optq.link_ids[3], lid1, "link_ids[3]");
 ASSERT_EQ(optq.prog_ids[4], 0, "prog_ids[4]");
 ASSERT_EQ(optq.link_ids[4], 0, "link_ids[4]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, true, "seen_tc3");
 ASSERT_EQ(skel->bss->seen_tc4, true, "seen_tc4");
cleanup:
 test_tc_link__destroy(skel);
 assert_mprog_count(target, 0);
}

void test_ns_tc_links_prepend(void)
{
 test_tc_links_prepend_target(BPF_TCX_INGRESS);
 test_tc_links_prepend_target(BPF_TCX_EGRESS);
}

static void test_tc_links_append_target(int target)
{
 LIBBPF_OPTS(bpf_prog_query_opts, optq);
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 prog_ids[5], link_ids[5];
 __u32 pid1, pid2, pid3, pid4;
 __u32 lid1, lid2, lid3, lid4;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err;

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc3, target),
    0, "tc3_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc4, target),
    0, "tc4_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));
 pid3 = id_from_prog_fd(bpf_program__fd(skel->progs.tc3));
 pid4 = id_from_prog_fd(bpf_program__fd(skel->progs.tc4));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");
 ASSERT_NEQ(pid3, pid4, "prog_ids_3_4");
 ASSERT_NEQ(pid2, pid3, "prog_ids_2_3");

 assert_mprog_count(target, 0);

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 lid1 = id_from_link_fd(bpf_link__fd(skel->links.tc1));

 assert_mprog_count(target, 1);

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_AFTER,
 );

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 lid2 = id_from_link_fd(bpf_link__fd(skel->links.tc2));

 assert_mprog_count(target, 2);

 optq.prog_ids = prog_ids;
 optq.link_ids = link_ids;

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 2, "count");
 ASSERT_EQ(optq.revision, 3, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid1, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid1, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid2, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid2, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], 0, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], 0, "link_ids[2]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, false, "seen_tc3");
 ASSERT_EQ(skel->bss->seen_tc4, false, "seen_tc4");

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_AFTER,
 );

 link = bpf_program__attach_tcx(skel->progs.tc3, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc3 = link;

 lid3 = id_from_link_fd(bpf_link__fd(skel->links.tc3));

 LIBBPF_OPTS_RESET(optl,
  .flags = BPF_F_AFTER,
 );

 link = bpf_program__attach_tcx(skel->progs.tc4, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc4 = link;

 lid4 = id_from_link_fd(bpf_link__fd(skel->links.tc4));

 assert_mprog_count(target, 4);

 memset(prog_ids, 0, sizeof(prog_ids));
 memset(link_ids, 0, sizeof(link_ids));
 optq.count = ARRAY_SIZE(prog_ids);

 err = bpf_prog_query_opts(loopback, target, &optq);
 if (!ASSERT_OK(err, "prog_query"))
  goto cleanup;

 ASSERT_EQ(optq.count, 4, "count");
 ASSERT_EQ(optq.revision, 5, "revision");
 ASSERT_EQ(optq.prog_ids[0], pid1, "prog_ids[0]");
 ASSERT_EQ(optq.link_ids[0], lid1, "link_ids[0]");
 ASSERT_EQ(optq.prog_ids[1], pid2, "prog_ids[1]");
 ASSERT_EQ(optq.link_ids[1], lid2, "link_ids[1]");
 ASSERT_EQ(optq.prog_ids[2], pid3, "prog_ids[2]");
 ASSERT_EQ(optq.link_ids[2], lid3, "link_ids[2]");
 ASSERT_EQ(optq.prog_ids[3], pid4, "prog_ids[3]");
 ASSERT_EQ(optq.link_ids[3], lid4, "link_ids[3]");
 ASSERT_EQ(optq.prog_ids[4], 0, "prog_ids[4]");
 ASSERT_EQ(optq.link_ids[4], 0, "link_ids[4]");

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, true, "seen_tc3");
 ASSERT_EQ(skel->bss->seen_tc4, true, "seen_tc4");
cleanup:
 test_tc_link__destroy(skel);
 assert_mprog_count(target, 0);
}

void test_ns_tc_links_append(void)
{
 test_tc_links_append_target(BPF_TCX_INGRESS);
 test_tc_links_append_target(BPF_TCX_EGRESS);
}

static void test_tc_links_dev_cleanup_target(int target)
{
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 LIBBPF_OPTS(bpf_prog_query_opts, optq);
 __u32 pid1, pid2, pid3, pid4;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err, ifindex;

 ASSERT_OK(system("ip link add dev tcx_opts1 type veth peer name tcx_opts2"), "add veth");
 ifindex = if_nametoindex("tcx_opts1");
 ASSERT_NEQ(ifindex, 0, "non_zero_ifindex");

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc3, target),
    0, "tc3_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc4, target),
    0, "tc4_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));
 pid3 = id_from_prog_fd(bpf_program__fd(skel->progs.tc3));
 pid4 = id_from_prog_fd(bpf_program__fd(skel->progs.tc4));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");
 ASSERT_NEQ(pid3, pid4, "prog_ids_3_4");
 ASSERT_NEQ(pid2, pid3, "prog_ids_2_3");

 assert_mprog_count(target, 0);

 link = bpf_program__attach_tcx(skel->progs.tc1, ifindex, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 assert_mprog_count_ifindex(ifindex, target, 1);

 link = bpf_program__attach_tcx(skel->progs.tc2, ifindex, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 assert_mprog_count_ifindex(ifindex, target, 2);

 link = bpf_program__attach_tcx(skel->progs.tc3, ifindex, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc3 = link;

 assert_mprog_count_ifindex(ifindex, target, 3);

 link = bpf_program__attach_tcx(skel->progs.tc4, ifindex, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc4 = link;

 assert_mprog_count_ifindex(ifindex, target, 4);

 ASSERT_OK(system("ip link del dev tcx_opts1"), "del veth");
 ASSERT_EQ(if_nametoindex("tcx_opts1"), 0, "dev1_removed");
 ASSERT_EQ(if_nametoindex("tcx_opts2"), 0, "dev2_removed");

 ASSERT_EQ(ifindex_from_link_fd(bpf_link__fd(skel->links.tc1)), 0, "tc1_ifindex");
 ASSERT_EQ(ifindex_from_link_fd(bpf_link__fd(skel->links.tc2)), 0, "tc2_ifindex");
 ASSERT_EQ(ifindex_from_link_fd(bpf_link__fd(skel->links.tc3)), 0, "tc3_ifindex");
 ASSERT_EQ(ifindex_from_link_fd(bpf_link__fd(skel->links.tc4)), 0, "tc4_ifindex");

 test_tc_link__destroy(skel);
 return;
cleanup:
 test_tc_link__destroy(skel);

 ASSERT_OK(system("ip link del dev tcx_opts1"), "del veth");
 ASSERT_EQ(if_nametoindex("tcx_opts1"), 0, "dev1_removed");
 ASSERT_EQ(if_nametoindex("tcx_opts2"), 0, "dev2_removed");
}

void test_ns_tc_links_dev_cleanup(void)
{
 test_tc_links_dev_cleanup_target(BPF_TCX_INGRESS);
 test_tc_links_dev_cleanup_target(BPF_TCX_EGRESS);
}

static void test_tc_chain_mixed(int target)
{
 LIBBPF_OPTS(bpf_tc_opts, tc_opts, .handle = 1, .priority = 1);
 LIBBPF_OPTS(bpf_tc_hook, tc_hook, .ifindex = loopback);
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 struct test_tc_link *skel;
 struct bpf_link *link;
 __u32 pid1, pid2, pid3;
 int err;

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc4, target),
    0, "tc4_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc5, target),
    0, "tc5_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc6, target),
    0, "tc6_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc4));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc5));
 pid3 = id_from_prog_fd(bpf_program__fd(skel->progs.tc6));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");
 ASSERT_NEQ(pid2, pid3, "prog_ids_2_3");

 assert_mprog_count(target, 0);

 tc_hook.attach_point = target == BPF_TCX_INGRESS ?
          BPF_TC_INGRESS : BPF_TC_EGRESS;
 err = bpf_tc_hook_create(&tc_hook);
 err = err == -EEXIST ? 0 : err;
 if (!ASSERT_OK(err, "bpf_tc_hook_create"))
  goto cleanup;

 tc_opts.prog_fd = bpf_program__fd(skel->progs.tc5);
 err = bpf_tc_attach(&tc_hook, &tc_opts);
 if (!ASSERT_OK(err, "bpf_tc_attach"))
  goto cleanup;

 link = bpf_program__attach_tcx(skel->progs.tc6, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc6 = link;

 assert_mprog_count(target, 1);

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc4, false, "seen_tc4");
 ASSERT_EQ(skel->bss->seen_tc5, false, "seen_tc5");
 ASSERT_EQ(skel->bss->seen_tc6, true, "seen_tc6");

 err = bpf_link__update_program(skel->links.tc6, skel->progs.tc4);
 if (!ASSERT_OK(err, "link_update"))
  goto cleanup;

 assert_mprog_count(target, 1);

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc4, true, "seen_tc4");
 ASSERT_EQ(skel->bss->seen_tc5, true, "seen_tc5");
 ASSERT_EQ(skel->bss->seen_tc6, false, "seen_tc6");

 err = bpf_link__detach(skel->links.tc6);
 if (!ASSERT_OK(err, "prog_detach"))
  goto cleanup;

 assert_mprog_count(target, 0);

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc4, false, "seen_tc4");
 ASSERT_EQ(skel->bss->seen_tc5, true, "seen_tc5");
 ASSERT_EQ(skel->bss->seen_tc6, false, "seen_tc6");

cleanup:
 tc_opts.flags = tc_opts.prog_fd = tc_opts.prog_id = 0;
 err = bpf_tc_detach(&tc_hook, &tc_opts);
 ASSERT_OK(err, "bpf_tc_detach");

 tc_hook.attach_point = BPF_TC_INGRESS | BPF_TC_EGRESS;
 bpf_tc_hook_destroy(&tc_hook);

 test_tc_link__destroy(skel);
}

void test_ns_tc_links_chain_mixed(void)
{
 test_tc_chain_mixed(BPF_TCX_INGRESS);
 test_tc_chain_mixed(BPF_TCX_EGRESS);
}

static void test_tc_links_ingress(int target, bool chain_tc_old,
      bool tcx_teardown_first)
{
 LIBBPF_OPTS(bpf_tc_opts, tc_opts,
  .handle  = 1,
  .priority = 1,
 );
 LIBBPF_OPTS(bpf_tc_hook, tc_hook,
  .ifindex = loopback,
  .attach_point = BPF_TC_CUSTOM,
  .parent  = TC_H_INGRESS,
 );
 bool hook_created = false, tc_attached = false;
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 pid1, pid2, pid3;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err;

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));
 pid3 = id_from_prog_fd(bpf_program__fd(skel->progs.tc3));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");
 ASSERT_NEQ(pid2, pid3, "prog_ids_2_3");

 assert_mprog_count(target, 0);

 if (chain_tc_old) {
  ASSERT_OK(system("tc qdisc add dev lo ingress"), "add_ingress");
  hook_created = true;

  tc_opts.prog_fd = bpf_program__fd(skel->progs.tc3);
  err = bpf_tc_attach(&tc_hook, &tc_opts);
  if (!ASSERT_OK(err, "bpf_tc_attach"))
   goto cleanup;
  tc_attached = true;
 }

 link = bpf_program__attach_tcx(skel->progs.tc1, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 link = bpf_program__attach_tcx(skel->progs.tc2, loopback, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 assert_mprog_count(target, 2);

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, true, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, chain_tc_old, "seen_tc3");

 err = bpf_link__detach(skel->links.tc2);
 if (!ASSERT_OK(err, "prog_detach"))
  goto cleanup;

 assert_mprog_count(target, 1);

 tc_skel_reset_all_seen(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);

 ASSERT_EQ(skel->bss->seen_tc1, true, "seen_tc1");
 ASSERT_EQ(skel->bss->seen_tc2, false, "seen_tc2");
 ASSERT_EQ(skel->bss->seen_tc3, chain_tc_old, "seen_tc3");
cleanup:
 if (tc_attached) {
  tc_opts.flags = tc_opts.prog_fd = tc_opts.prog_id = 0;
  err = bpf_tc_detach(&tc_hook, &tc_opts);
  ASSERT_OK(err, "bpf_tc_detach");
 }
 ASSERT_OK(system(ping_cmd), ping_cmd);
 assert_mprog_count(target, 1);
 if (hook_created && tcx_teardown_first)
  ASSERT_OK(system("tc qdisc del dev lo ingress"), "del_ingress");
 ASSERT_OK(system(ping_cmd), ping_cmd);
 test_tc_link__destroy(skel);
 ASSERT_OK(system(ping_cmd), ping_cmd);
 if (hook_created && !tcx_teardown_first)
  ASSERT_OK(system("tc qdisc del dev lo ingress"), "del_ingress");
 ASSERT_OK(system(ping_cmd), ping_cmd);
 assert_mprog_count(target, 0);
}

void test_ns_tc_links_ingress(void)
{
 test_tc_links_ingress(BPF_TCX_INGRESS, true, true);
 test_tc_links_ingress(BPF_TCX_INGRESS, true, false);
 test_tc_links_ingress(BPF_TCX_INGRESS, false, false);
}

struct qdisc_req {
 struct nlmsghdr  n;
 struct tcmsg     t;
 char             buf[1024];
};

static int qdisc_replace(int ifindex, const char *kind, bool block)
{
 struct rtnl_handle rth = { .fd = -1 };
 struct qdisc_req req;
 int err;

 err = rtnl_open(&rth, 0);
 if (!ASSERT_OK(err, "open_rtnetlink"))
  return err;

 memset(&req, 0, sizeof(req));
 req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct tcmsg));
 req.n.nlmsg_flags = NLM_F_CREATE | NLM_F_REPLACE | NLM_F_REQUEST;
 req.n.nlmsg_type = RTM_NEWQDISC;
 req.t.tcm_family = AF_UNSPEC;
 req.t.tcm_ifindex = ifindex;
 req.t.tcm_parent = 0xfffffff1;

 addattr_l(&req.n, sizeof(req), TCA_KIND, kind, strlen(kind) + 1);
 if (block)
  addattr32(&req.n, sizeof(req), TCA_INGRESS_BLOCK, 1);

 err = rtnl_talk(&rth, &req.n, NULL);
 ASSERT_OK(err, "talk_rtnetlink");
 rtnl_close(&rth);
 return err;
}

void test_ns_tc_links_dev_chain0(void)
{
 int err, ifindex;

 ASSERT_OK(system("ip link add dev foo type veth peer name bar"), "add veth");
 ifindex = if_nametoindex("foo");
 ASSERT_NEQ(ifindex, 0, "non_zero_ifindex");
 err = qdisc_replace(ifindex, "ingress", true);
 if (!ASSERT_OK(err, "attaching ingress"))
  goto cleanup;
 ASSERT_OK(system("tc filter add block 1 matchall action skbmod swap mac"), "add block");
 err = qdisc_replace(ifindex, "clsact", false);
 if (!ASSERT_OK(err, "attaching clsact"))
  goto cleanup;
 /* Heuristic: kern_sync_rcu() alone does not work; a wait-time of ~5s
 * triggered the issue without the fix reliably 100% of the time.
 */
 sleep(5);
 ASSERT_OK(system("tc filter add dev foo ingress matchall action skbmod swap mac"), "add filter");
cleanup:
 ASSERT_OK(system("ip link del dev foo"), "del veth");
 ASSERT_EQ(if_nametoindex("foo"), 0, "foo removed");
 ASSERT_EQ(if_nametoindex("bar"), 0, "bar removed");
}

static void test_tc_links_dev_mixed(int target)
{
 LIBBPF_OPTS(bpf_tc_opts, tc_opts, .handle = 1, .priority = 1);
 LIBBPF_OPTS(bpf_tc_hook, tc_hook);
 LIBBPF_OPTS(bpf_tcx_opts, optl);
 __u32 pid1, pid2, pid3, pid4;
 struct test_tc_link *skel;
 struct bpf_link *link;
 int err, ifindex;

 ASSERT_OK(system("ip link add dev tcx_opts1 type veth peer name tcx_opts2"), "add veth");
 ifindex = if_nametoindex("tcx_opts1");
 ASSERT_NEQ(ifindex, 0, "non_zero_ifindex");

 skel = test_tc_link__open();
 if (!ASSERT_OK_PTR(skel, "skel_open"))
  goto cleanup;

 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc1, target),
    0, "tc1_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc2, target),
    0, "tc2_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc3, target),
    0, "tc3_attach_type");
 ASSERT_EQ(bpf_program__set_expected_attach_type(skel->progs.tc4, target),
    0, "tc4_attach_type");

 err = test_tc_link__load(skel);
 if (!ASSERT_OK(err, "skel_load"))
  goto cleanup;

 pid1 = id_from_prog_fd(bpf_program__fd(skel->progs.tc1));
 pid2 = id_from_prog_fd(bpf_program__fd(skel->progs.tc2));
 pid3 = id_from_prog_fd(bpf_program__fd(skel->progs.tc3));
 pid4 = id_from_prog_fd(bpf_program__fd(skel->progs.tc4));

 ASSERT_NEQ(pid1, pid2, "prog_ids_1_2");
 ASSERT_NEQ(pid3, pid4, "prog_ids_3_4");
 ASSERT_NEQ(pid2, pid3, "prog_ids_2_3");

 assert_mprog_count(target, 0);

 link = bpf_program__attach_tcx(skel->progs.tc1, ifindex, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc1 = link;

 assert_mprog_count_ifindex(ifindex, target, 1);

 link = bpf_program__attach_tcx(skel->progs.tc2, ifindex, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc2 = link;

 assert_mprog_count_ifindex(ifindex, target, 2);

 link = bpf_program__attach_tcx(skel->progs.tc3, ifindex, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc3 = link;

 assert_mprog_count_ifindex(ifindex, target, 3);

 link = bpf_program__attach_tcx(skel->progs.tc4, ifindex, &optl);
 if (!ASSERT_OK_PTR(link, "link_attach"))
  goto cleanup;

 skel->links.tc4 = link;

 assert_mprog_count_ifindex(ifindex, target, 4);

 tc_hook.ifindex = ifindex;
 tc_hook.attach_point = target == BPF_TCX_INGRESS ?
          BPF_TC_INGRESS : BPF_TC_EGRESS;

 err = bpf_tc_hook_create(&tc_hook);
 err = err == -EEXIST ? 0 : err;
 if (!ASSERT_OK(err, "bpf_tc_hook_create"))
  goto cleanup;

 tc_opts.prog_fd = bpf_program__fd(skel->progs.tc5);
 err = bpf_tc_attach(&tc_hook, &tc_opts);
 if (!ASSERT_OK(err, "bpf_tc_attach"))
  goto cleanup;

 ASSERT_OK(system("ip link del dev tcx_opts1"), "del veth");
 ASSERT_EQ(if_nametoindex("tcx_opts1"), 0, "dev1_removed");
 ASSERT_EQ(if_nametoindex("tcx_opts2"), 0, "dev2_removed");

 ASSERT_EQ(ifindex_from_link_fd(bpf_link__fd(skel->links.tc1)), 0, "tc1_ifindex");
 ASSERT_EQ(ifindex_from_link_fd(bpf_link__fd(skel->links.tc2)), 0, "tc2_ifindex");
 ASSERT_EQ(ifindex_from_link_fd(bpf_link__fd(skel->links.tc3)), 0, "tc3_ifindex");
 ASSERT_EQ(ifindex_from_link_fd(bpf_link__fd(skel->links.tc4)), 0, "tc4_ifindex");

 test_tc_link__destroy(skel);
 return;
cleanup:
 test_tc_link__destroy(skel);

 ASSERT_OK(system("ip link del dev tcx_opts1"), "del veth");
 ASSERT_EQ(if_nametoindex("tcx_opts1"), 0, "dev1_removed");
 ASSERT_EQ(if_nametoindex("tcx_opts2"), 0, "dev2_removed");
}

void test_ns_tc_links_dev_mixed(void)
{
 test_tc_links_dev_mixed(BPF_TCX_INGRESS);
 test_tc_links_dev_mixed(BPF_TCX_EGRESS);
}