| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* Copyright (c) 2022 Google */ |
| 3 | |
| 4 | #include <test_progs.h> |
| 5 | #include <bpf/libbpf.h> |
| 6 | #include <bpf/btf.h> |
| 7 | #include "iters_css_task.skel.h" |
| 8 | #include "cgroup_iter.skel.h" |
| 9 | #include "cgroup_helpers.h" |
| 10 | |
| 11 | #define ROOT 0 |
| 12 | #define PARENT 1 |
| 13 | #define CHILD1 2 |
| 14 | #define CHILD2 3 |
| 15 | #define NUM_CGROUPS 4 |
| 16 | |
| 17 | #define PROLOGUE "prologue\n" |
| 18 | #define EPILOGUE "epilogue\n" |
| 19 | |
| 20 | static const char *cg_path[] = { |
| 21 | "/" , "/parent" , "/parent/child1" , "/parent/child2" |
| 22 | }; |
| 23 | |
| 24 | static int cg_fd[] = {-1, -1, -1, -1}; |
| 25 | static unsigned long long cg_id[] = {0, 0, 0, 0}; |
| 26 | static char expected_output[64]; |
| 27 | |
| 28 | static int setup_cgroups(void) |
| 29 | { |
| 30 | int fd, i = 0; |
| 31 | |
| 32 | for (i = 0; i < NUM_CGROUPS; i++) { |
| 33 | fd = create_and_get_cgroup(cg_path[i]); |
| 34 | if (fd < 0) |
| 35 | return fd; |
| 36 | |
| 37 | cg_fd[i] = fd; |
| 38 | cg_id[i] = get_cgroup_id(cg_path[i]); |
| 39 | } |
| 40 | return 0; |
| 41 | } |
| 42 | |
| 43 | static void cleanup_cgroups(void) |
| 44 | { |
| 45 | int i; |
| 46 | |
| 47 | for (i = 0; i < NUM_CGROUPS; i++) |
| 48 | close(cg_fd[i]); |
| 49 | } |
| 50 | |
| 51 | static void read_from_cgroup_iter(struct bpf_program *prog, int cgroup_fd, |
| 52 | int order, const char *testname) |
| 53 | { |
| 54 | DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts); |
| 55 | union bpf_iter_link_info linfo; |
| 56 | struct bpf_link *link; |
| 57 | int len, iter_fd; |
| 58 | static char buf[128]; |
| 59 | size_t left; |
| 60 | char *p; |
| 61 | |
| 62 | memset(&linfo, 0, sizeof(linfo)); |
| 63 | linfo.cgroup.cgroup_fd = cgroup_fd; |
| 64 | linfo.cgroup.order = order; |
| 65 | opts.link_info = &linfo; |
| 66 | opts.link_info_len = sizeof(linfo); |
| 67 | |
| 68 | link = bpf_program__attach_iter(prog, &opts); |
| 69 | if (!ASSERT_OK_PTR(link, "attach_iter" )) |
| 70 | return; |
| 71 | |
| 72 | iter_fd = bpf_iter_create(bpf_link__fd(link)); |
| 73 | if (iter_fd < 0) |
| 74 | goto free_link; |
| 75 | |
| 76 | memset(buf, 0, sizeof(buf)); |
| 77 | left = ARRAY_SIZE(buf); |
| 78 | p = buf; |
| 79 | while ((len = read(iter_fd, p, left)) > 0) { |
| 80 | p += len; |
| 81 | left -= len; |
| 82 | } |
| 83 | |
| 84 | ASSERT_STREQ(buf, expected_output, testname); |
| 85 | |
| 86 | /* read() after iter finishes should be ok. */ |
| 87 | if (len == 0) |
| 88 | ASSERT_OK(read(iter_fd, buf, sizeof(buf)), "second_read" ); |
| 89 | |
| 90 | close(iter_fd); |
| 91 | free_link: |
| 92 | bpf_link__destroy(link); |
| 93 | } |
| 94 | |
| 95 | /* Invalid cgroup. */ |
| 96 | static void test_invalid_cgroup(struct cgroup_iter *skel) |
| 97 | { |
| 98 | DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts); |
| 99 | union bpf_iter_link_info linfo; |
| 100 | struct bpf_link *link; |
| 101 | |
| 102 | memset(&linfo, 0, sizeof(linfo)); |
| 103 | linfo.cgroup.cgroup_fd = (__u32)-1; |
| 104 | opts.link_info = &linfo; |
| 105 | opts.link_info_len = sizeof(linfo); |
| 106 | |
| 107 | link = bpf_program__attach_iter(skel->progs.cgroup_id_printer, &opts); |
| 108 | ASSERT_ERR_PTR(link, "attach_iter" ); |
| 109 | bpf_link__destroy(link); |
| 110 | } |
| 111 | |
| 112 | /* Specifying both cgroup_fd and cgroup_id is invalid. */ |
| 113 | static void test_invalid_cgroup_spec(struct cgroup_iter *skel) |
| 114 | { |
| 115 | DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts); |
| 116 | union bpf_iter_link_info linfo; |
| 117 | struct bpf_link *link; |
| 118 | |
| 119 | memset(&linfo, 0, sizeof(linfo)); |
| 120 | linfo.cgroup.cgroup_fd = (__u32)cg_fd[PARENT]; |
| 121 | linfo.cgroup.cgroup_id = (__u64)cg_id[PARENT]; |
| 122 | opts.link_info = &linfo; |
| 123 | opts.link_info_len = sizeof(linfo); |
| 124 | |
| 125 | link = bpf_program__attach_iter(skel->progs.cgroup_id_printer, &opts); |
| 126 | ASSERT_ERR_PTR(link, "attach_iter" ); |
| 127 | bpf_link__destroy(link); |
| 128 | } |
| 129 | |
| 130 | /* Preorder walk prints parent and child in order. */ |
| 131 | static void test_walk_preorder(struct cgroup_iter *skel) |
| 132 | { |
| 133 | snprintf(expected_output, sizeof(expected_output), |
| 134 | PROLOGUE "%8llu\n%8llu\n%8llu\n" EPILOGUE, |
| 135 | cg_id[PARENT], cg_id[CHILD1], cg_id[CHILD2]); |
| 136 | |
| 137 | read_from_cgroup_iter(prog: skel->progs.cgroup_id_printer, cgroup_fd: cg_fd[PARENT], |
| 138 | order: BPF_CGROUP_ITER_DESCENDANTS_PRE, testname: "preorder" ); |
| 139 | } |
| 140 | |
| 141 | /* Postorder walk prints child and parent in order. */ |
| 142 | static void test_walk_postorder(struct cgroup_iter *skel) |
| 143 | { |
| 144 | snprintf(expected_output, sizeof(expected_output), |
| 145 | PROLOGUE "%8llu\n%8llu\n%8llu\n" EPILOGUE, |
| 146 | cg_id[CHILD1], cg_id[CHILD2], cg_id[PARENT]); |
| 147 | |
| 148 | read_from_cgroup_iter(prog: skel->progs.cgroup_id_printer, cgroup_fd: cg_fd[PARENT], |
| 149 | order: BPF_CGROUP_ITER_DESCENDANTS_POST, testname: "postorder" ); |
| 150 | } |
| 151 | |
| 152 | /* Walking parents prints parent and then root. */ |
| 153 | static void test_walk_ancestors_up(struct cgroup_iter *skel) |
| 154 | { |
| 155 | /* terminate the walk when ROOT is met. */ |
| 156 | skel->bss->terminal_cgroup = cg_id[ROOT]; |
| 157 | |
| 158 | snprintf(expected_output, sizeof(expected_output), |
| 159 | PROLOGUE "%8llu\n%8llu\n" EPILOGUE, |
| 160 | cg_id[PARENT], cg_id[ROOT]); |
| 161 | |
| 162 | read_from_cgroup_iter(prog: skel->progs.cgroup_id_printer, cgroup_fd: cg_fd[PARENT], |
| 163 | order: BPF_CGROUP_ITER_ANCESTORS_UP, testname: "ancestors_up" ); |
| 164 | |
| 165 | skel->bss->terminal_cgroup = 0; |
| 166 | } |
| 167 | |
| 168 | /* Early termination prints parent only. */ |
| 169 | static void test_early_termination(struct cgroup_iter *skel) |
| 170 | { |
| 171 | /* terminate the walk after the first element is processed. */ |
| 172 | skel->bss->terminate_early = 1; |
| 173 | |
| 174 | snprintf(expected_output, sizeof(expected_output), |
| 175 | PROLOGUE "%8llu\n" EPILOGUE, cg_id[PARENT]); |
| 176 | |
| 177 | read_from_cgroup_iter(prog: skel->progs.cgroup_id_printer, cgroup_fd: cg_fd[PARENT], |
| 178 | order: BPF_CGROUP_ITER_DESCENDANTS_PRE, testname: "early_termination" ); |
| 179 | |
| 180 | skel->bss->terminate_early = 0; |
| 181 | } |
| 182 | |
| 183 | /* Waling self prints self only. */ |
| 184 | static void test_walk_self_only(struct cgroup_iter *skel) |
| 185 | { |
| 186 | snprintf(expected_output, sizeof(expected_output), |
| 187 | PROLOGUE "%8llu\n" EPILOGUE, cg_id[PARENT]); |
| 188 | |
| 189 | read_from_cgroup_iter(prog: skel->progs.cgroup_id_printer, cgroup_fd: cg_fd[PARENT], |
| 190 | order: BPF_CGROUP_ITER_SELF_ONLY, testname: "self_only" ); |
| 191 | } |
| 192 | |
| 193 | static void test_walk_dead_self_only(struct cgroup_iter *skel) |
| 194 | { |
| 195 | DECLARE_LIBBPF_OPTS(bpf_iter_attach_opts, opts); |
| 196 | char expected_output[128], buf[128]; |
| 197 | const char *cgrp_name = "/dead" ; |
| 198 | union bpf_iter_link_info linfo; |
| 199 | int len, cgrp_fd, iter_fd; |
| 200 | struct bpf_link *link; |
| 201 | size_t left; |
| 202 | char *p; |
| 203 | |
| 204 | cgrp_fd = create_and_get_cgroup(cgrp_name); |
| 205 | if (!ASSERT_GE(cgrp_fd, 0, "create cgrp" )) |
| 206 | return; |
| 207 | |
| 208 | /* The cgroup will be dead during read() iteration, so it only has |
| 209 | * epilogue in the output |
| 210 | */ |
| 211 | snprintf(expected_output, sizeof(expected_output), EPILOGUE); |
| 212 | |
| 213 | memset(&linfo, 0, sizeof(linfo)); |
| 214 | linfo.cgroup.cgroup_fd = cgrp_fd; |
| 215 | linfo.cgroup.order = BPF_CGROUP_ITER_SELF_ONLY; |
| 216 | opts.link_info = &linfo; |
| 217 | opts.link_info_len = sizeof(linfo); |
| 218 | |
| 219 | link = bpf_program__attach_iter(skel->progs.cgroup_id_printer, &opts); |
| 220 | if (!ASSERT_OK_PTR(link, "attach_iter" )) |
| 221 | goto close_cgrp; |
| 222 | |
| 223 | iter_fd = bpf_iter_create(bpf_link__fd(link)); |
| 224 | if (!ASSERT_GE(iter_fd, 0, "iter_create" )) |
| 225 | goto free_link; |
| 226 | |
| 227 | /* Close link fd and cgroup fd */ |
| 228 | bpf_link__destroy(link); |
| 229 | close(cgrp_fd); |
| 230 | |
| 231 | /* Remove cgroup to mark it as dead */ |
| 232 | remove_cgroup(cgrp_name); |
| 233 | |
| 234 | /* Two kern_sync_rcu() and usleep() pairs are used to wait for the |
| 235 | * releases of cgroup css, and the last kern_sync_rcu() and usleep() |
| 236 | * pair is used to wait for the free of cgroup itself. |
| 237 | */ |
| 238 | kern_sync_rcu(); |
| 239 | usleep(8000); |
| 240 | kern_sync_rcu(); |
| 241 | usleep(8000); |
| 242 | kern_sync_rcu(); |
| 243 | usleep(1000); |
| 244 | |
| 245 | memset(buf, 0, sizeof(buf)); |
| 246 | left = ARRAY_SIZE(buf); |
| 247 | p = buf; |
| 248 | while ((len = read(iter_fd, p, left)) > 0) { |
| 249 | p += len; |
| 250 | left -= len; |
| 251 | } |
| 252 | |
| 253 | ASSERT_STREQ(buf, expected_output, "dead cgroup output" ); |
| 254 | |
| 255 | /* read() after iter finishes should be ok. */ |
| 256 | if (len == 0) |
| 257 | ASSERT_OK(read(iter_fd, buf, sizeof(buf)), "second_read" ); |
| 258 | |
| 259 | close(iter_fd); |
| 260 | return; |
| 261 | free_link: |
| 262 | bpf_link__destroy(link); |
| 263 | close_cgrp: |
| 264 | close(cgrp_fd); |
| 265 | } |
| 266 | |
| 267 | static void test_walk_self_only_css_task(void) |
| 268 | { |
| 269 | struct iters_css_task *skel; |
| 270 | int err; |
| 271 | |
| 272 | skel = iters_css_task__open(); |
| 273 | if (!ASSERT_OK_PTR(skel, "skel_open" )) |
| 274 | return; |
| 275 | |
| 276 | bpf_program__set_autoload(skel->progs.cgroup_id_printer, true); |
| 277 | |
| 278 | err = iters_css_task__load(skel); |
| 279 | if (!ASSERT_OK(err, "skel_load" )) |
| 280 | goto cleanup; |
| 281 | |
| 282 | err = join_cgroup(cg_path[CHILD2]); |
| 283 | if (!ASSERT_OK(err, "join_cgroup" )) |
| 284 | goto cleanup; |
| 285 | |
| 286 | skel->bss->target_pid = getpid(); |
| 287 | snprintf(expected_output, sizeof(expected_output), |
| 288 | PROLOGUE "%8llu\n" EPILOGUE, cg_id[CHILD2]); |
| 289 | read_from_cgroup_iter(skel->progs.cgroup_id_printer, cg_fd[CHILD2], |
| 290 | BPF_CGROUP_ITER_SELF_ONLY, "test_walk_self_only_css_task" ); |
| 291 | ASSERT_EQ(skel->bss->css_task_cnt, 1, "css_task_cnt" ); |
| 292 | cleanup: |
| 293 | iters_css_task__destroy(skel); |
| 294 | } |
| 295 | |
| 296 | void test_cgroup_iter(void) |
| 297 | { |
| 298 | struct cgroup_iter *skel = NULL; |
| 299 | |
| 300 | if (setup_cgroup_environment()) |
| 301 | return; |
| 302 | |
| 303 | if (setup_cgroups()) |
| 304 | goto out; |
| 305 | |
| 306 | skel = cgroup_iter__open_and_load(); |
| 307 | if (!ASSERT_OK_PTR(skel, "cgroup_iter__open_and_load" )) |
| 308 | goto out; |
| 309 | |
| 310 | if (test__start_subtest("cgroup_iter__invalid_cgroup" )) |
| 311 | test_invalid_cgroup(skel); |
| 312 | if (test__start_subtest("cgroup_iter__invalid_cgroup_spec" )) |
| 313 | test_invalid_cgroup_spec(skel); |
| 314 | if (test__start_subtest("cgroup_iter__preorder" )) |
| 315 | test_walk_preorder(skel); |
| 316 | if (test__start_subtest("cgroup_iter__postorder" )) |
| 317 | test_walk_postorder(skel); |
| 318 | if (test__start_subtest("cgroup_iter__ancestors_up_walk" )) |
| 319 | test_walk_ancestors_up(skel); |
| 320 | if (test__start_subtest("cgroup_iter__early_termination" )) |
| 321 | test_early_termination(skel); |
| 322 | if (test__start_subtest("cgroup_iter__self_only" )) |
| 323 | test_walk_self_only(skel); |
| 324 | if (test__start_subtest("cgroup_iter__dead_self_only" )) |
| 325 | test_walk_dead_self_only(skel); |
| 326 | if (test__start_subtest("cgroup_iter__self_only_css_task" )) |
| 327 | test_walk_self_only_css_task(); |
| 328 | |
| 329 | out: |
| 330 | cgroup_iter__destroy(skel); |
| 331 | cleanup_cgroups(); |
| 332 | cleanup_cgroup_environment(); |
| 333 | } |
| 334 | |