| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | #include "ebpf.h" |
| 4 | #include "ebpf_process.h" |
| 5 | #include "libbpf_api/ebpf_library.h" |
| 6 | |
| 7 | /***************************************************************** |
| 8 | * |
| 9 | * GLOBAL VARIABLES |
| 10 | * |
| 11 | *****************************************************************/ |
| 12 | |
| 13 | static char *process_dimension_names[NETDATA_KEY_PUBLISH_PROCESS_END] = {"process", "task", "process", "thread"}; |
| 14 | static char *process_id_names[NETDATA_KEY_PUBLISH_PROCESS_END] = {"do_exit", "release_task", "_do_fork", "sys_clone"}; |
| 15 | static char *status[] = {"process", "zombie"}; |
| 16 | |
| 17 | netdata_ebpf_targets_t process_targets[] = { |
| 18 | {.name = "release_task", .mode = EBPF_LOAD_TRAMPOLINE}, |
| 19 | {.name = "__x64_sys_clone", .mode = EBPF_LOAD_TRAMPOLINE}, |
| 20 | {.name = "__x64_sys_clone3", .mode = EBPF_LOAD_TRAMPOLINE}, |
| 21 | {.name = "_do_fork", .mode = EBPF_LOAD_TRAMPOLINE}, |
| 22 | {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; |
| 23 | |
| 24 | static ebpf_local_maps_t process_maps[] = { |
| 25 | {.name = "tbl_pid_stats", |
| 26 | .internal_input = ND_EBPF_DEFAULT_PID_SIZE, |
| 27 | .user_input = 0, |
| 28 | .type = NETDATA_EBPF_MAP_RESIZABLE | NETDATA_EBPF_MAP_PID, |
| 29 | .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED, |
| 30 | #ifdef LIBBPF_MAJOR_VERSION |
| 31 | .map_type = BPF_MAP_TYPE_PERCPU_HASH |
| 32 | #endif |
| 33 | }, |
| 34 | {.name = "tbl_total_stats", |
| 35 | .internal_input = NETDATA_KEY_END_VECTOR, |
| 36 | .user_input = 0, |
| 37 | .type = NETDATA_EBPF_MAP_STATIC, |
| 38 | .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED, |
| 39 | #ifdef LIBBPF_MAJOR_VERSION |
| 40 | .map_type = BPF_MAP_TYPE_PERCPU_ARRAY |
| 41 | #endif |
| 42 | }, |
| 43 | {.name = "process_ctrl", |
| 44 | .internal_input = NETDATA_CONTROLLER_END, |
| 45 | .user_input = 0, |
| 46 | .type = NETDATA_EBPF_MAP_CONTROLLER, |
| 47 | .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED, |
| 48 | #ifdef LIBBPF_MAJOR_VERSION |
| 49 | .map_type = BPF_MAP_TYPE_PERCPU_ARRAY |
| 50 | #endif |
| 51 | }, |
| 52 | {.name = NULL, |
| 53 | .internal_input = 0, |
| 54 | .user_input = 0, |
| 55 | .type = NETDATA_EBPF_MAP_CONTROLLER, |
| 56 | .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED, |
| 57 | #ifdef LIBBPF_MAJOR_VERSION |
| 58 | .map_type = BPF_MAP_TYPE_PERCPU_ARRAY |
| 59 | #endif |
| 60 | }}; |
| 61 | |
| 62 | char *tracepoint_sched_type = "sched"; |
| 63 | char *tracepoint_sched_process_exit = "sched_process_exit"; |
| 64 | char *tracepoint_sched_process_exec = "sched_process_exec"; |
| 65 | char *tracepoint_sched_process_fork = "sched_process_fork"; |
| 66 | static int was_sched_process_exit_enabled = 0; |
| 67 | static int was_sched_process_exec_enabled = 0; |
| 68 | static int was_sched_process_fork_enabled = 0; |
| 69 | |
| 70 | static netdata_idx_t *process_hash_values = NULL; |
| 71 | ebpf_process_stat_t *process_stat_vector = NULL; |
| 72 | static bool process_safe_clean = false; |
| 73 | static netdata_syscall_stat_t process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_END]; |
| 74 | static netdata_publish_syscall_t process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_END]; |
| 75 | |
| 76 | struct config process_config = APPCONFIG_INITIALIZER; |
| 77 | |
| 78 | #ifdef LIBBPF_MAJOR_VERSION |
| 79 | /** |
| 80 | * Disable probe |
| 81 | * |
| 82 | * Disable all probes to use exclusively another method. |
| 83 | * |
| 84 | * @param obj is the main structure for bpf objects |
| 85 | */ |
| 86 | static void ebpf_process_disable_probe(struct process_bpf *obj) |
| 87 | { |
| 88 | bpf_program__set_autoload(obj->progs.netdata_release_task_probe, false); |
| 89 | bpf_program__set_autoload(obj->progs.netdata_do_fork_probe, false); |
| 90 | bpf_program__set_autoload(obj->progs.netdata_kernel_clone_probe, false); |
| 91 | } |
| 92 | |
| 93 | static void ebpf_disable_tracepoints(struct process_bpf *obj) |
| 94 | { |
| 95 | bpf_program__set_autoload(obj->progs.netdata_clone_exit, false); |
| 96 | bpf_program__set_autoload(obj->progs.netdata_clone3_exit, false); |
| 97 | bpf_program__set_autoload(obj->progs.netdata_fork_exit, false); |
| 98 | bpf_program__set_autoload(obj->progs.netdata_vfork_exit, false); |
| 99 | } |
| 100 | |
| 101 | static void ebpf_set_trampoline_target(struct process_bpf *obj) |
| 102 | { |
| 103 | bpf_program__set_attach_target( |
| 104 | obj->progs.netdata_release_task_fentry, 0, process_targets[PROCESS_RELEASE_TASK_NAME].name); |
| 105 | |
| 106 | bpf_program__set_attach_target(obj->progs.netdata_clone_fexit, 0, process_targets[PROCESS_SYS_CLONE].name); |
| 107 | |
| 108 | bpf_program__set_attach_target(obj->progs.netdata_clone3_fexit, 0, process_targets[PROCESS_SYS_CLONE3].name); |
| 109 | } |
| 110 | |
| 111 | /* |
| 112 | * Disable trampoline |
| 113 | * |
| 114 | * Disable all trampoline to use exclusively another method. |
| 115 | * |
| 116 | * @param obj is the main structure for bpf objects. |
| 117 | */ |
| 118 | static void ebpf_disable_trampoline(struct process_bpf *obj) |
| 119 | { |
| 120 | bpf_program__set_autoload(obj->progs.netdata_release_task_fentry, false); |
| 121 | bpf_program__set_autoload(obj->progs.netdata_clone_fexit, false); |
| 122 | bpf_program__set_autoload(obj->progs.netdata_clone3_fexit, false); |
| 123 | } |
| 124 | |
| 125 | static inline void ebpf_disable_clone3(struct process_bpf *obj) |
| 126 | { |
| 127 | bpf_program__set_autoload(obj->progs.netdata_clone3_exit, false); |
| 128 | bpf_program__set_autoload(obj->progs.netdata_clone3_fexit, false); |
| 129 | } |
| 130 | |
| 131 | static inline void ebpf_adjust_process_fork(struct process_bpf *obj) |
| 132 | { |
| 133 | if (running_on_kernel <= NETDATA_EBPF_KERNEL_6_16) { |
| 134 | bpf_program__set_autoload(obj->progs.netdata_tracepoint_sched_process_fork, true); |
| 135 | bpf_program__set_autoload(obj->progs.netdata_tracepoint_sched_process_fork_v2, false); |
| 136 | } else { |
| 137 | bpf_program__set_autoload(obj->progs.netdata_tracepoint_sched_process_fork_v2, true); |
| 138 | bpf_program__set_autoload(obj->progs.netdata_tracepoint_sched_process_fork, false); |
| 139 | } |
| 140 | } |
| 141 | |
| 142 | /** |
| 143 | * Mount Attach Probe |
| 144 | * |
| 145 | * Attach probes to target |
| 146 | * |
| 147 | * @param obj is the main structure for bpf objects. |
| 148 | * |
| 149 | * @return It returns 0 on success and -1 otherwise. |
| 150 | */ |
| 151 | static inline int process_attach_kprobe_target(struct process_bpf *obj) |
| 152 | { |
| 153 | obj->links.netdata_release_task_probe = bpf_program__attach_kprobe( |
| 154 | obj->progs.netdata_release_task_probe, false, process_targets[PROCESS_RELEASE_TASK_NAME].name); |
| 155 | int ret = libbpf_get_error(obj->links.netdata_release_task_probe); |
| 156 | if (ret) |
| 157 | goto endakt; |
| 158 | |
| 159 | if (running_on_kernel < NETDATA_EBPF_KERNEL_5_9_16) { |
| 160 | obj->links.netdata_do_fork_probe = |
| 161 | bpf_program__attach_kprobe(obj->progs.netdata_do_fork_probe, false, process_targets[PROCESS_SYS_FORK].name); |
| 162 | ret = libbpf_get_error(obj->links.netdata_do_fork_probe); |
| 163 | } else { |
| 164 | obj->links.netdata_kernel_clone_probe = bpf_program__attach_kprobe( |
| 165 | obj->progs.netdata_kernel_clone_probe, false, process_targets[PROCESS_KERNEL_CLONE].name); |
| 166 | ret = libbpf_get_error(obj->links.netdata_kernel_clone_probe); |
| 167 | } |
| 168 | endakt: |
| 169 | return ret; |
| 170 | } |
| 171 | |
| 172 | /** |
| 173 | * Set hash tables |
| 174 | * |
| 175 | * Set the values for maps according the value given by kernel. |
| 176 | * |
| 177 | * @param obj is the main structure for bpf objects. |
| 178 | */ |
| 179 | static void ebpf_process_set_hash_tables(struct process_bpf *obj) |
| 180 | { |
| 181 | process_maps[NETDATA_PROCESS_GLOBAL_TABLE].map_fd = bpf_map__fd(obj->maps.tbl_total_stats); |
| 182 | process_maps[NETDATA_PROCESS_PID_TABLE].map_fd = bpf_map__fd(obj->maps.tbl_pid_stats); |
| 183 | process_maps[NETDATA_PROCESS_CTRL_TABLE].map_fd = bpf_map__fd(obj->maps.process_ctrl); |
| 184 | } |
| 185 | |
| 186 | /** |
| 187 | * Load and attach |
| 188 | * |
| 189 | * Load and attach the eBPF code in kernel. |
| 190 | * |
| 191 | * @param obj is the main structure for bpf objects. |
| 192 | * @param em structure with configuration |
| 193 | * |
| 194 | * @return it returns 0 on success and -1 otherwise |
| 195 | */ |
| 196 | static inline int ebpf_process_load_and_attach(struct process_bpf *obj, ebpf_module_t *em) |
| 197 | { |
| 198 | netdata_ebpf_program_loaded_t mode = em->targets[PROCESS_RELEASE_TASK_NAME].mode; |
| 199 | if (mode == EBPF_LOAD_TRAMPOLINE) { |
| 200 | ebpf_process_disable_probe(obj); |
| 201 | ebpf_disable_tracepoints(obj); |
| 202 | |
| 203 | ebpf_set_trampoline_target(obj); |
| 204 | } else if (mode == EBPF_LOAD_PROBE || mode == EBPF_LOAD_RETPROBE) { |
| 205 | ebpf_disable_tracepoints(obj); |
| 206 | ebpf_disable_trampoline(obj); |
| 207 | |
| 208 | bpf_program__set_autoload( |
| 209 | (running_on_kernel <= NETDATA_EBPF_KERNEL_5_9_16) ? obj->progs.netdata_kernel_clone_probe : |
| 210 | obj->progs.netdata_do_fork_probe, |
| 211 | false); |
| 212 | } else { // Tracepoint |
| 213 | ebpf_process_disable_probe(obj); |
| 214 | ebpf_disable_trampoline(obj); |
| 215 | } |
| 216 | |
| 217 | if (running_on_kernel < NETDATA_EBPF_KERNEL_5_3) { |
| 218 | ebpf_disable_clone3(obj); |
| 219 | } |
| 220 | |
| 221 | ebpf_adjust_process_fork(obj); |
| 222 | |
| 223 | int ret = process_bpf__load(obj); |
| 224 | if (ret) { |
| 225 | return ret; |
| 226 | } |
| 227 | |
| 228 | ret = (mode == EBPF_LOAD_TRAMPOLINE) ? process_bpf__attach(obj) : process_attach_kprobe_target(obj); |
| 229 | if (!ret) { |
| 230 | ebpf_process_set_hash_tables(obj); |
| 231 | |
| 232 | ebpf_update_controller(cachestat_maps[NETDATA_PROCESS_CTRL_TABLE].map_fd, em); |
| 233 | } |
| 234 | |
| 235 | return ret; |
| 236 | } |
| 237 | #endif |
| 238 | |
| 239 | /* |
| 240 | * Load BPF |
| 241 | * |
| 242 | * Load BPF files. |
| 243 | * |
| 244 | * @param em the structure with configuration |
| 245 | */ |
| 246 | static int ebpf_process_load_bpf(ebpf_module_t *em) |
| 247 | { |
| 248 | #ifdef LIBBPF_MAJOR_VERSION |
| 249 | ebpf_define_map_type(process_maps, em->maps_per_core, running_on_kernel); |
| 250 | #endif |
| 251 | |
| 252 | int ret = 0; |
| 253 | // ebpf_adjust_apps_cgroup(em, em->targets[NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU].mode); |
| 254 | if (em->load & EBPF_LOAD_LEGACY) { |
| 255 | em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); |
| 256 | if (!em->probe_links) { |
| 257 | ret = -1; |
| 258 | } |
| 259 | } |
| 260 | #ifdef LIBBPF_MAJOR_VERSION |
| 261 | else { |
| 262 | process_bpf_obj = process_bpf__open(); |
| 263 | if (!process_bpf_obj) |
| 264 | ret = -1; |
| 265 | else { |
| 266 | ret = ebpf_process_load_and_attach(process_bpf_obj, em); |
| 267 | if (ret) { |
| 268 | process_bpf__destroy(process_bpf_obj); |
| 269 | process_bpf_obj = NULL; |
| 270 | } |
| 271 | } |
| 272 | } |
| 273 | #endif |
| 274 | |
| 275 | if (ret) |
| 276 | netdata_log_error("%s %s", EBPF_DEFAULT_ERROR_MSG, em->info.thread_name); |
| 277 | |
| 278 | return ret; |
| 279 | } |
| 280 | |
| 281 | /***************************************************************** |
| 282 | * |
| 283 | * PROCESS DATA AND SEND TO NETDATA |
| 284 | * |
| 285 | *****************************************************************/ |
| 286 | |
| 287 | /** |
| 288 | * Update publish structure before to send data to Netdata. |
| 289 | * |
| 290 | * @param publish the first output structure with independent dimensions |
| 291 | * @param pvc the second output structure with correlated dimensions |
| 292 | * @param input the structure with the input data. |
| 293 | */ |
| 294 | static void ebpf_update_global_publish( |
| 295 | netdata_publish_syscall_t *publish, |
| 296 | netdata_publish_vfs_common_t *pvc, |
| 297 | netdata_syscall_stat_t *input) |
| 298 | { |
| 299 | netdata_publish_syscall_t *move = publish; |
| 300 | int selector = NETDATA_KEY_PUBLISH_PROCESS_EXIT; |
| 301 | while (move) { |
| 302 | move->ncall = (input->call > move->pcall) ? input->call - move->pcall : move->pcall - input->call; |
| 303 | move->nbyte = (input->bytes > move->pbyte) ? input->bytes - move->pbyte : move->pbyte - input->bytes; |
| 304 | move->nerr = (input->ecall > move->nerr) ? input->ecall - move->perr : move->perr - input->ecall; |
| 305 | |
| 306 | move->pcall = input->call; |
| 307 | move->pbyte = input->bytes; |
| 308 | move->perr = input->ecall; |
| 309 | |
| 310 | input = input->next; |
| 311 | move = move->next; |
| 312 | selector++; |
| 313 | } |
| 314 | |
| 315 | pvc->running = |
| 316 | (long)publish[NETDATA_KEY_PUBLISH_PROCESS_FORK].ncall - (long)publish[NETDATA_KEY_PUBLISH_PROCESS_CLONE].ncall; |
| 317 | pvc->zombie = (long)publish[NETDATA_KEY_PUBLISH_PROCESS_EXIT].ncall - |
| 318 | (long)publish[NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK].ncall; |
| 319 | } |
| 320 | |
| 321 | /** |
| 322 | * Call the necessary functions to create a chart. |
| 323 | * |
| 324 | * @param family the chart family |
| 325 | * @param move the pointer with the values that will be published |
| 326 | */ |
| 327 | static void write_status_chart(char *family, netdata_publish_vfs_common_t *pvc) |
| 328 | { |
| 329 | ebpf_write_begin_chart(family, NETDATA_PROCESS_STATUS_NAME, ""); |
| 330 | |
| 331 | write_chart_dimension(status[0], (long long)pvc->running); |
| 332 | write_chart_dimension(status[1], (long long)pvc->zombie); |
| 333 | |
| 334 | ebpf_write_end_chart(); |
| 335 | } |
| 336 | |
| 337 | /** |
| 338 | * Send data to Netdata calling auxiliary functions. |
| 339 | * |
| 340 | * @param em the structure with thread information |
| 341 | */ |
| 342 | static void ebpf_process_send_data(ebpf_module_t *em) |
| 343 | { |
| 344 | netdata_publish_vfs_common_t pvc; |
| 345 | ebpf_update_global_publish(process_publish_aggregated, &pvc, process_aggregated_data); |
| 346 | |
| 347 | write_count_chart( |
| 348 | NETDATA_EXIT_SYSCALL, |
| 349 | NETDATA_EBPF_SYSTEM_GROUP, |
| 350 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT], |
| 351 | 2); |
| 352 | write_count_chart( |
| 353 | NETDATA_PROCESS_SYSCALL, |
| 354 | NETDATA_EBPF_SYSTEM_GROUP, |
| 355 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK], |
| 356 | 2); |
| 357 | |
| 358 | write_status_chart(NETDATA_EBPF_SYSTEM_GROUP, &pvc); |
| 359 | if (em->mode < MODE_ENTRY) { |
| 360 | write_err_chart( |
| 361 | NETDATA_PROCESS_ERROR_NAME, |
| 362 | NETDATA_EBPF_SYSTEM_GROUP, |
| 363 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK], |
| 364 | 2); |
| 365 | } |
| 366 | } |
| 367 | |
| 368 | /** |
| 369 | * Send data to Netdata calling auxiliary functions. |
| 370 | * |
| 371 | * @param root the target list. |
| 372 | */ |
| 373 | void ebpf_process_send_apps_data(struct ebpf_target *root, ebpf_module_t *em) |
| 374 | { |
| 375 | struct ebpf_target *w; |
| 376 | |
| 377 | for (w = root; w; w = w->next) { |
| 378 | if (ebpf_plugin_stop()) |
| 379 | break; |
| 380 | |
| 381 | if (unlikely(!(w->charts_created & (1 << EBPF_MODULE_PROCESS_IDX)))) |
| 382 | continue; |
| 383 | |
| 384 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, w->clean_name, "_ebpf_process_start"); |
| 385 | write_chart_dimension("calls", w->process.create_process); |
| 386 | ebpf_write_end_chart(); |
| 387 | |
| 388 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, w->clean_name, "_ebpf_thread_start"); |
| 389 | write_chart_dimension("calls", w->process.create_thread); |
| 390 | ebpf_write_end_chart(); |
| 391 | |
| 392 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, w->clean_name, "_ebpf_task_exit"); |
| 393 | write_chart_dimension("calls", w->process.exit_call); |
| 394 | ebpf_write_end_chart(); |
| 395 | |
| 396 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, w->clean_name, "_ebpf_task_released"); |
| 397 | write_chart_dimension("calls", w->process.release_call); |
| 398 | ebpf_write_end_chart(); |
| 399 | |
| 400 | if (em->mode < MODE_ENTRY) { |
| 401 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, w->clean_name, "_ebpf_task_error"); |
| 402 | write_chart_dimension("calls", w->process.task_err); |
| 403 | ebpf_write_end_chart(); |
| 404 | } |
| 405 | } |
| 406 | } |
| 407 | |
| 408 | /***************************************************************** |
| 409 | * |
| 410 | * READ INFORMATION FROM KERNEL RING |
| 411 | * |
| 412 | *****************************************************************/ |
| 413 | |
| 414 | /** |
| 415 | * Read the hash table and store data to allocated vectors. |
| 416 | * |
| 417 | * @param maps_per_core do I need to read all cores? |
| 418 | */ |
| 419 | static void ebpf_read_process_hash_global_tables(netdata_idx_t *stats, int maps_per_core) |
| 420 | { |
| 421 | netdata_idx_t res[NETDATA_KEY_END_VECTOR]; |
| 422 | ebpf_read_global_table_stats( |
| 423 | res, |
| 424 | process_hash_values, |
| 425 | process_maps[NETDATA_PROCESS_GLOBAL_TABLE].map_fd, |
| 426 | maps_per_core, |
| 427 | 0, |
| 428 | NETDATA_KEY_END_VECTOR); |
| 429 | |
| 430 | ebpf_read_global_table_stats( |
| 431 | stats, |
| 432 | process_hash_values, |
| 433 | process_maps[NETDATA_PROCESS_CTRL_TABLE].map_fd, |
| 434 | maps_per_core, |
| 435 | NETDATA_CONTROLLER_PID_TABLE_ADD, |
| 436 | NETDATA_CONTROLLER_END); |
| 437 | |
| 438 | process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_EXIT].call = res[NETDATA_KEY_CALLS_DO_EXIT]; |
| 439 | process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK].call = res[NETDATA_KEY_CALLS_RELEASE_TASK]; |
| 440 | process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_FORK].call = res[NETDATA_KEY_CALLS_DO_FORK]; |
| 441 | process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_CLONE].call = res[NETDATA_KEY_CALLS_SYS_CLONE]; |
| 442 | |
| 443 | process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_FORK].ecall = res[NETDATA_KEY_ERROR_DO_FORK]; |
| 444 | process_aggregated_data[NETDATA_KEY_PUBLISH_PROCESS_CLONE].ecall = res[NETDATA_KEY_ERROR_SYS_CLONE]; |
| 445 | } |
| 446 | |
| 447 | /** |
| 448 | * Update cgroup |
| 449 | * |
| 450 | * Update cgroup data based in PID running. |
| 451 | * |
| 452 | * @param maps_per_core do I need to read all cores? |
| 453 | */ |
| 454 | static void ebpf_update_process_cgroup() |
| 455 | { |
| 456 | ebpf_cgroup_target_t *ect; |
| 457 | netdata_mutex_lock(&mutex_cgroup_shm); |
| 458 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 459 | if (ebpf_plugin_stop()) |
| 460 | break; |
| 461 | |
| 462 | struct pid_on_target2 *pids; |
| 463 | for (pids = ect->pids; pids; pids = pids->next) { |
| 464 | uint32_t pid = pids->pid; |
| 465 | ebpf_publish_process_t *out = &pids->ps; |
| 466 | netdata_ebpf_pid_stats_t *local_pid = netdata_ebpf_lookup_shm_pointer_unsafe(pid); |
| 467 | if (!local_pid || !(local_pid->threads & (1U << (NETDATA_EBPF_PIDS_PROCESS_IDX << 1)))) |
| 468 | continue; |
| 469 | |
| 470 | ebpf_publish_process_t *in = &local_pid->process; |
| 471 | |
| 472 | memcpy(out, in, sizeof(ebpf_publish_process_t)); |
| 473 | } |
| 474 | } |
| 475 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 476 | } |
| 477 | |
| 478 | /***************************************************************** |
| 479 | * |
| 480 | * FUNCTIONS TO CREATE CHARTS |
| 481 | * |
| 482 | *****************************************************************/ |
| 483 | |
| 484 | /** |
| 485 | * Create process status chart |
| 486 | * |
| 487 | * @param family the chart family |
| 488 | * @param name the chart name |
| 489 | * @param axis the axis label |
| 490 | * @param web the group name used to attach the chart on dashboard |
| 491 | * @param order the order number of the specified chart |
| 492 | * @param update_every value to overwrite the update frequency set by the server. |
| 493 | */ |
| 494 | static void |
| 495 | ebpf_process_status_chart(char *family, char *name, char *axis, char *web, char *algorithm, int order, int update_every) |
| 496 | { |
| 497 | printf( |
| 498 | "CHART %s.%s '' 'Process not closed' '%s' '%s' 'system.process_status' line %d %d '' 'ebpf.plugin' 'process'\n", |
| 499 | family, |
| 500 | name, |
| 501 | axis, |
| 502 | web, |
| 503 | order, |
| 504 | update_every); |
| 505 | |
| 506 | printf("DIMENSION %s '' %s 1 1\n", status[0], algorithm); |
| 507 | printf("DIMENSION %s '' %s 1 1\n", status[1], algorithm); |
| 508 | } |
| 509 | |
| 510 | /** |
| 511 | * Create global charts |
| 512 | * |
| 513 | * Call ebpf_create_chart to create the charts for the collector. |
| 514 | * |
| 515 | * @param em a pointer to the structure with the default values. |
| 516 | */ |
| 517 | static void ebpf_create_global_charts(ebpf_module_t *em) |
| 518 | { |
| 519 | ebpf_create_chart( |
| 520 | NETDATA_EBPF_SYSTEM_GROUP, |
| 521 | NETDATA_PROCESS_SYSCALL, |
| 522 | "Start process", |
| 523 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 524 | NETDATA_PROCESS_GROUP, |
| 525 | "system.process_thread", |
| 526 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 527 | 21002, |
| 528 | ebpf_create_global_dimension, |
| 529 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK], |
| 530 | 2, |
| 531 | em->update_every, |
| 532 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 533 | |
| 534 | ebpf_create_chart( |
| 535 | NETDATA_EBPF_SYSTEM_GROUP, |
| 536 | NETDATA_EXIT_SYSCALL, |
| 537 | "Exit process", |
| 538 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 539 | NETDATA_PROCESS_GROUP, |
| 540 | "system.exit", |
| 541 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 542 | 21003, |
| 543 | ebpf_create_global_dimension, |
| 544 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT], |
| 545 | 2, |
| 546 | em->update_every, |
| 547 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 548 | |
| 549 | ebpf_process_status_chart( |
| 550 | NETDATA_EBPF_SYSTEM_GROUP, |
| 551 | NETDATA_PROCESS_STATUS_NAME, |
| 552 | EBPF_COMMON_UNITS_CALLS, |
| 553 | NETDATA_PROCESS_GROUP, |
| 554 | ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX], |
| 555 | 21004, |
| 556 | em->update_every); |
| 557 | |
| 558 | if (em->mode < MODE_ENTRY) { |
| 559 | ebpf_create_chart( |
| 560 | NETDATA_EBPF_SYSTEM_GROUP, |
| 561 | NETDATA_PROCESS_ERROR_NAME, |
| 562 | "Fails to create process", |
| 563 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 564 | NETDATA_PROCESS_GROUP, |
| 565 | "system.task_error", |
| 566 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 567 | 21005, |
| 568 | ebpf_create_global_dimension, |
| 569 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK], |
| 570 | 2, |
| 571 | em->update_every, |
| 572 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 573 | } |
| 574 | |
| 575 | fflush(stdout); |
| 576 | } |
| 577 | |
| 578 | /** |
| 579 | * Create process apps charts |
| 580 | * |
| 581 | * Call ebpf_create_chart to create the charts on apps submenu. |
| 582 | * |
| 583 | * @param em a pointer to the structure with the default values. |
| 584 | * @param ptr a pointer for the targets. |
| 585 | */ |
| 586 | void ebpf_process_create_apps_charts(struct ebpf_module *em, void *ptr) |
| 587 | { |
| 588 | struct ebpf_target *root = ptr; |
| 589 | struct ebpf_target *w; |
| 590 | int update_every = em->update_every; |
| 591 | for (w = root; w; w = w->next) { |
| 592 | if (ebpf_plugin_stop()) |
| 593 | break; |
| 594 | |
| 595 | if (unlikely(!w->exposed)) |
| 596 | continue; |
| 597 | |
| 598 | ebpf_write_chart_cmd( |
| 599 | NETDATA_APP_FAMILY, |
| 600 | w->clean_name, |
| 601 | "_ebpf_process_start", |
| 602 | "Process started.", |
| 603 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 604 | NETDATA_PROCESS_GROUP, |
| 605 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 606 | "app.ebpf_process_start", |
| 607 | 20161, |
| 608 | update_every, |
| 609 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 610 | ebpf_create_chart_labels("app_group", w->name, RRDLABEL_SRC_AUTO); |
| 611 | ebpf_commit_label(); |
| 612 | fprintf(stdout, "DIMENSION calls '' %s 1 1\n", ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); |
| 613 | |
| 614 | ebpf_write_chart_cmd( |
| 615 | NETDATA_APP_FAMILY, |
| 616 | w->clean_name, |
| 617 | "_ebpf_thread_start", |
| 618 | "Threads started.", |
| 619 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 620 | NETDATA_PROCESS_GROUP, |
| 621 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 622 | "app.ebpf_thread_start", |
| 623 | 20162, |
| 624 | update_every, |
| 625 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 626 | ebpf_create_chart_labels("app_group", w->name, RRDLABEL_SRC_AUTO); |
| 627 | ebpf_commit_label(); |
| 628 | fprintf(stdout, "DIMENSION calls '' %s 1 1\n", ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); |
| 629 | |
| 630 | ebpf_write_chart_cmd( |
| 631 | NETDATA_APP_FAMILY, |
| 632 | w->clean_name, |
| 633 | "_ebpf_task_exit", |
| 634 | "Tasks starts exit process.", |
| 635 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 636 | NETDATA_PROCESS_GROUP, |
| 637 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 638 | "app.ebpf_task_exit", |
| 639 | 20163, |
| 640 | update_every, |
| 641 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 642 | ebpf_create_chart_labels("app_group", w->name, RRDLABEL_SRC_AUTO); |
| 643 | ebpf_commit_label(); |
| 644 | fprintf(stdout, "DIMENSION calls '' %s 1 1\n", ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); |
| 645 | |
| 646 | ebpf_write_chart_cmd( |
| 647 | NETDATA_APP_FAMILY, |
| 648 | w->clean_name, |
| 649 | "_ebpf_task_released", |
| 650 | "Tasks released.", |
| 651 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 652 | NETDATA_PROCESS_GROUP, |
| 653 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 654 | "app.ebpf_task_released", |
| 655 | 20164, |
| 656 | update_every, |
| 657 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 658 | ebpf_create_chart_labels("app_group", w->name, RRDLABEL_SRC_AUTO); |
| 659 | ebpf_commit_label(); |
| 660 | fprintf(stdout, "DIMENSION calls '' %s 1 1\n", ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); |
| 661 | |
| 662 | if (em->mode < MODE_ENTRY) { |
| 663 | ebpf_write_chart_cmd( |
| 664 | NETDATA_APP_FAMILY, |
| 665 | w->clean_name, |
| 666 | "_ebpf_task_error", |
| 667 | "Errors to create process or threads.", |
| 668 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 669 | NETDATA_PROCESS_GROUP, |
| 670 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 671 | "app.ebpf_task_error", |
| 672 | 20165, |
| 673 | update_every, |
| 674 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 675 | ebpf_create_chart_labels("app_group", w->name, RRDLABEL_SRC_AUTO); |
| 676 | ebpf_commit_label(); |
| 677 | fprintf(stdout, "DIMENSION calls '' %s 1 1\n", ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); |
| 678 | } |
| 679 | w->charts_created |= 1 << EBPF_MODULE_PROCESS_IDX; |
| 680 | } |
| 681 | |
| 682 | em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; |
| 683 | } |
| 684 | |
| 685 | /***************************************************************** |
| 686 | * |
| 687 | * FUNCTIONS TO CLOSE THE THREAD |
| 688 | * |
| 689 | *****************************************************************/ |
| 690 | |
| 691 | static void ebpf_obsolete_specific_process_charts(char *type, ebpf_module_t *em); |
| 692 | |
| 693 | /** |
| 694 | * Obsolete services |
| 695 | * |
| 696 | * Obsolete all service charts created |
| 697 | * |
| 698 | * @param em a pointer to `struct ebpf_module` |
| 699 | */ |
| 700 | static void ebpf_obsolete_process_services(ebpf_module_t *em, char *id) |
| 701 | { |
| 702 | ebpf_write_chart_obsolete( |
| 703 | id, |
| 704 | NETDATA_SYSCALL_APPS_TASK_PROCESS, |
| 705 | "", |
| 706 | "Process started", |
| 707 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 708 | NETDATA_APPS_PROCESS_GROUP, |
| 709 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 710 | NETDATA_SYSTEMD_PROCESS_CREATE_CONTEXT, |
| 711 | 20065, |
| 712 | em->update_every); |
| 713 | |
| 714 | ebpf_write_chart_obsolete( |
| 715 | id, |
| 716 | NETDATA_SYSCALL_APPS_TASK_THREAD, |
| 717 | "", |
| 718 | "Threads started", |
| 719 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 720 | NETDATA_APPS_PROCESS_GROUP, |
| 721 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 722 | NETDATA_SYSTEMD_THREAD_CREATE_CONTEXT, |
| 723 | 20066, |
| 724 | em->update_every); |
| 725 | |
| 726 | ebpf_write_chart_obsolete( |
| 727 | id, |
| 728 | NETDATA_SYSCALL_APPS_TASK_CLOSE, |
| 729 | "", |
| 730 | "Tasks starts exit process.", |
| 731 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 732 | NETDATA_APPS_PROCESS_GROUP, |
| 733 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 734 | NETDATA_SYSTEMD_PROCESS_EXIT_CONTEXT, |
| 735 | 20067, |
| 736 | em->update_every); |
| 737 | |
| 738 | ebpf_write_chart_obsolete( |
| 739 | id, |
| 740 | NETDATA_SYSCALL_APPS_TASK_EXIT, |
| 741 | "", |
| 742 | "Tasks closed", |
| 743 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 744 | NETDATA_APPS_PROCESS_GROUP, |
| 745 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 746 | NETDATA_SYSTEMD_PROCESS_CLOSE_CONTEXT, |
| 747 | 20068, |
| 748 | em->update_every); |
| 749 | |
| 750 | if (em->mode < MODE_ENTRY) { |
| 751 | ebpf_write_chart_obsolete( |
| 752 | id, |
| 753 | NETDATA_SYSCALL_APPS_TASK_ERROR, |
| 754 | "", |
| 755 | "Errors to create process or threads.", |
| 756 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 757 | NETDATA_APPS_PROCESS_GROUP, |
| 758 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 759 | NETDATA_SYSTEMD_PROCESS_ERROR_CONTEXT, |
| 760 | 20069, |
| 761 | em->update_every); |
| 762 | } |
| 763 | } |
| 764 | |
| 765 | /** |
| 766 | * Obsolete cgroup chart |
| 767 | * |
| 768 | * Send obsolete for all charts created before to close. |
| 769 | * |
| 770 | * @param em a pointer to `struct ebpf_module` |
| 771 | */ |
| 772 | static inline void ebpf_obsolete_process_cgroup_charts(ebpf_module_t *em) |
| 773 | { |
| 774 | netdata_mutex_lock(&mutex_cgroup_shm); |
| 775 | |
| 776 | ebpf_cgroup_target_t *ect; |
| 777 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 778 | if (ect->systemd) { |
| 779 | ebpf_obsolete_process_services(em, ect->name); |
| 780 | |
| 781 | continue; |
| 782 | } |
| 783 | |
| 784 | ebpf_obsolete_specific_process_charts(ect->name, em); |
| 785 | } |
| 786 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 787 | } |
| 788 | |
| 789 | /** |
| 790 | * Obsolette apps charts |
| 791 | * |
| 792 | * Obsolete apps charts. |
| 793 | * |
| 794 | * @param em a pointer to the structure with the default values. |
| 795 | */ |
| 796 | void ebpf_obsolete_process_apps_charts(struct ebpf_module *em) |
| 797 | { |
| 798 | struct ebpf_target *w; |
| 799 | int update_every = em->update_every; |
| 800 | for (w = apps_groups_root_target; w; w = w->next) { |
| 801 | if (unlikely(!(w->charts_created & (1 << EBPF_MODULE_PROCESS_IDX)))) |
| 802 | continue; |
| 803 | |
| 804 | ebpf_write_chart_obsolete( |
| 805 | NETDATA_APP_FAMILY, |
| 806 | w->clean_name, |
| 807 | "_ebpf_process_start", |
| 808 | "Process started.", |
| 809 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 810 | NETDATA_PROCESS_GROUP, |
| 811 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 812 | "app.ebpf_process_start", |
| 813 | 20161, |
| 814 | update_every); |
| 815 | |
| 816 | ebpf_write_chart_obsolete( |
| 817 | NETDATA_APP_FAMILY, |
| 818 | w->clean_name, |
| 819 | "_ebpf_thread_start", |
| 820 | "Threads started.", |
| 821 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 822 | NETDATA_PROCESS_GROUP, |
| 823 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 824 | "app.ebpf_thread_start", |
| 825 | 20162, |
| 826 | update_every); |
| 827 | |
| 828 | ebpf_write_chart_obsolete( |
| 829 | NETDATA_APP_FAMILY, |
| 830 | w->clean_name, |
| 831 | "_ebpf_task_exit", |
| 832 | "Tasks starts exit process.", |
| 833 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 834 | NETDATA_PROCESS_GROUP, |
| 835 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 836 | "app.ebpf_task_exit", |
| 837 | 20163, |
| 838 | update_every); |
| 839 | |
| 840 | ebpf_write_chart_obsolete( |
| 841 | NETDATA_APP_FAMILY, |
| 842 | w->clean_name, |
| 843 | "_ebpf_task_released", |
| 844 | "Tasks released.", |
| 845 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 846 | NETDATA_PROCESS_GROUP, |
| 847 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 848 | "app.ebpf_task_released", |
| 849 | 20164, |
| 850 | update_every); |
| 851 | |
| 852 | if (em->mode < MODE_ENTRY) { |
| 853 | ebpf_write_chart_obsolete( |
| 854 | NETDATA_APP_FAMILY, |
| 855 | w->clean_name, |
| 856 | "_ebpf_task_error", |
| 857 | "Errors to create process or threads.", |
| 858 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 859 | NETDATA_PROCESS_GROUP, |
| 860 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 861 | "app.ebpf_task_error", |
| 862 | 20165, |
| 863 | update_every); |
| 864 | } |
| 865 | |
| 866 | w->charts_created &= ~(1 << EBPF_MODULE_PROCESS_IDX); |
| 867 | } |
| 868 | } |
| 869 | |
| 870 | /** |
| 871 | * Obsolete global |
| 872 | * |
| 873 | * Obsolete global charts created by thread. |
| 874 | * |
| 875 | * @param em a pointer to `struct ebpf_module` |
| 876 | */ |
| 877 | static void ebpf_obsolete_process_global(ebpf_module_t *em) |
| 878 | { |
| 879 | ebpf_write_chart_obsolete( |
| 880 | NETDATA_EBPF_SYSTEM_GROUP, |
| 881 | NETDATA_PROCESS_SYSCALL, |
| 882 | "", |
| 883 | "Start process", |
| 884 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 885 | NETDATA_PROCESS_GROUP, |
| 886 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 887 | "system.process_thread", |
| 888 | 21002, |
| 889 | em->update_every); |
| 890 | |
| 891 | ebpf_write_chart_obsolete( |
| 892 | NETDATA_EBPF_SYSTEM_GROUP, |
| 893 | NETDATA_EXIT_SYSCALL, |
| 894 | "", |
| 895 | "Exit process", |
| 896 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 897 | NETDATA_PROCESS_GROUP, |
| 898 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 899 | "system.exit", |
| 900 | 21003, |
| 901 | em->update_every); |
| 902 | |
| 903 | ebpf_write_chart_obsolete( |
| 904 | NETDATA_EBPF_SYSTEM_GROUP, |
| 905 | NETDATA_PROCESS_STATUS_NAME, |
| 906 | "", |
| 907 | "Process not closed", |
| 908 | EBPF_COMMON_UNITS_CALLS, |
| 909 | NETDATA_PROCESS_GROUP, |
| 910 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 911 | "system.process_status", |
| 912 | 21004, |
| 913 | em->update_every); |
| 914 | |
| 915 | if (em->mode < MODE_ENTRY) { |
| 916 | ebpf_write_chart_obsolete( |
| 917 | NETDATA_EBPF_SYSTEM_GROUP, |
| 918 | NETDATA_PROCESS_ERROR_NAME, |
| 919 | "", |
| 920 | "Fails to create process", |
| 921 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 922 | NETDATA_PROCESS_GROUP, |
| 923 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 924 | "system.task_error", |
| 925 | 21005, |
| 926 | em->update_every); |
| 927 | } |
| 928 | } |
| 929 | |
| 930 | /** |
| 931 | * Process disable tracepoints |
| 932 | * |
| 933 | * Disable tracepoints when the plugin was responsible to enable it. |
| 934 | */ |
| 935 | static void ebpf_process_disable_tracepoints() |
| 936 | { |
| 937 | char *default_message = "Cannot disable the tracepoint"; |
| 938 | if (!was_sched_process_exit_enabled) { |
| 939 | if (ebpf_disable_tracing_values(tracepoint_sched_type, tracepoint_sched_process_exit)) |
| 940 | netdata_log_error("%s %s/%s.", default_message, tracepoint_sched_type, tracepoint_sched_process_exit); |
| 941 | } |
| 942 | |
| 943 | if (!was_sched_process_exec_enabled) { |
| 944 | if (ebpf_disable_tracing_values(tracepoint_sched_type, tracepoint_sched_process_exec)) |
| 945 | netdata_log_error("%s %s/%s.", default_message, tracepoint_sched_type, tracepoint_sched_process_exec); |
| 946 | } |
| 947 | |
| 948 | if (!was_sched_process_fork_enabled) { |
| 949 | if (ebpf_disable_tracing_values(tracepoint_sched_type, tracepoint_sched_process_fork)) |
| 950 | netdata_log_error("%s %s/%s.", default_message, tracepoint_sched_type, tracepoint_sched_process_fork); |
| 951 | } |
| 952 | } |
| 953 | |
| 954 | static void ebpf_process_exit(void *pptr) |
| 955 | { |
| 956 | ebpf_set_pid_map_fd(NETDATA_EBPF_PIDS_PROCESS_IDX, -1); |
| 957 | ebpf_module_t *em = CLEANUP_FUNCTION_GET_PTR(pptr); |
| 958 | if (!em) |
| 959 | return; |
| 960 | |
| 961 | if (!process_safe_clean) { |
| 962 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 963 | ebpf_module_enabled_set(em, NETDATA_THREAD_EBPF_STOPPED); |
| 964 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 965 | return; |
| 966 | } |
| 967 | |
| 968 | netdata_mutex_lock(&lock); |
| 969 | collect_pids &= ~(1 << EBPF_MODULE_PROCESS_IDX); |
| 970 | netdata_mutex_unlock(&lock); |
| 971 | |
| 972 | // Drop this module's bits from the shared PID pool so its slots don't |
| 973 | // stay pinned if the plugin keeps running after the module stops. |
| 974 | if (integration_shm && ebpf_shm_sem_wait_or_stop(shm_mutex_ebpf_integration)) { |
| 975 | netdata_ebpf_sweep_shm_for_module_unsafe(NETDATA_EBPF_PIDS_PROCESS_IDX); |
| 976 | sem_post(shm_mutex_ebpf_integration); |
| 977 | } |
| 978 | |
| 979 | if (ebpf_module_enabled_get(em) == NETDATA_THREAD_EBPF_FUNCTION_RUNNING && !ebpf_plugin_stop()) { |
| 980 | netdata_mutex_lock(&lock); |
| 981 | if (em->cgroup_charts) { |
| 982 | ebpf_obsolete_process_cgroup_charts(em); |
| 983 | fflush(stdout); |
| 984 | } |
| 985 | |
| 986 | if (em->apps_charts & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) { |
| 987 | ebpf_obsolete_process_apps_charts(em); |
| 988 | } |
| 989 | |
| 990 | ebpf_obsolete_process_global(em); |
| 991 | |
| 992 | fflush(stdout); |
| 993 | netdata_mutex_unlock(&lock); |
| 994 | } |
| 995 | |
| 996 | freez(process_hash_values); |
| 997 | freez(process_stat_vector); |
| 998 | |
| 999 | ebpf_process_disable_tracepoints(); |
| 1000 | |
| 1001 | if (!ebpf_plugin_stop() && em->functions.bpf_unload) |
| 1002 | em->functions.bpf_unload(em); |
| 1003 | |
| 1004 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 1005 | process_pid_fd = -1; |
| 1006 | ebpf_module_enabled_set(em, NETDATA_THREAD_EBPF_STOPPED); |
| 1007 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 1008 | } |
| 1009 | |
| 1010 | /***************************************************************** |
| 1011 | * |
| 1012 | * FUNCTIONS WITH THE MAIN LOOP |
| 1013 | * |
| 1014 | *****************************************************************/ |
| 1015 | |
| 1016 | /** |
| 1017 | * Sum PIDs |
| 1018 | * |
| 1019 | * Sum values for all targets. |
| 1020 | * |
| 1021 | * @param ps structure used to store data |
| 1022 | * @param pids input data |
| 1023 | */ |
| 1024 | static void ebpf_process_sum_cgroup_pids(ebpf_publish_process_t *ps, struct pid_on_target2 *pids) |
| 1025 | { |
| 1026 | ebpf_publish_process_t accumulator; |
| 1027 | memset(&accumulator, 0, sizeof(accumulator)); |
| 1028 | |
| 1029 | while (pids) { |
| 1030 | ebpf_publish_process_t *pps = &pids->ps; |
| 1031 | |
| 1032 | accumulator.exit_call += pps->exit_call; |
| 1033 | accumulator.release_call += pps->release_call; |
| 1034 | accumulator.create_process += pps->create_process; |
| 1035 | accumulator.create_thread += pps->create_thread; |
| 1036 | |
| 1037 | accumulator.task_err += pps->task_err; |
| 1038 | |
| 1039 | pids = pids->next; |
| 1040 | } |
| 1041 | |
| 1042 | ps->exit_call = (accumulator.exit_call >= ps->exit_call) ? accumulator.exit_call : ps->exit_call; |
| 1043 | ps->release_call = (accumulator.release_call >= ps->release_call) ? accumulator.release_call : ps->release_call; |
| 1044 | ps->create_process = |
| 1045 | (accumulator.create_process >= ps->create_process) ? accumulator.create_process : ps->create_process; |
| 1046 | ps->create_thread = |
| 1047 | (accumulator.create_thread >= ps->create_thread) ? accumulator.create_thread : ps->create_thread; |
| 1048 | |
| 1049 | ps->task_err = (accumulator.task_err >= ps->task_err) ? accumulator.task_err : ps->task_err; |
| 1050 | } |
| 1051 | |
| 1052 | /* |
| 1053 | * Send Specific Process data |
| 1054 | * |
| 1055 | * Send data for specific cgroup/apps. |
| 1056 | * |
| 1057 | * @param type chart type |
| 1058 | * @param values structure with values that will be sent to netdata |
| 1059 | * @param em the structure with thread information |
| 1060 | */ |
| 1061 | static void ebpf_send_specific_process_data(char *type, ebpf_publish_process_t *values, ebpf_module_t *em) |
| 1062 | { |
| 1063 | ebpf_write_begin_chart(type, NETDATA_SYSCALL_APPS_TASK_PROCESS, ""); |
| 1064 | write_chart_dimension( |
| 1065 | process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK].name, (long long)values->create_process); |
| 1066 | ebpf_write_end_chart(); |
| 1067 | |
| 1068 | ebpf_write_begin_chart(type, NETDATA_SYSCALL_APPS_TASK_THREAD, ""); |
| 1069 | write_chart_dimension( |
| 1070 | process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_CLONE].name, (long long)values->create_thread); |
| 1071 | ebpf_write_end_chart(); |
| 1072 | |
| 1073 | ebpf_write_begin_chart(type, NETDATA_SYSCALL_APPS_TASK_EXIT, ""); |
| 1074 | write_chart_dimension( |
| 1075 | process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT].name, (long long)values->release_call); |
| 1076 | ebpf_write_end_chart(); |
| 1077 | |
| 1078 | ebpf_write_begin_chart(type, NETDATA_SYSCALL_APPS_TASK_CLOSE, ""); |
| 1079 | write_chart_dimension( |
| 1080 | process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK].name, (long long)values->release_call); |
| 1081 | ebpf_write_end_chart(); |
| 1082 | |
| 1083 | if (em->mode < MODE_ENTRY) { |
| 1084 | ebpf_write_begin_chart(type, NETDATA_SYSCALL_APPS_TASK_ERROR, ""); |
| 1085 | write_chart_dimension( |
| 1086 | process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT].name, (long long)values->task_err); |
| 1087 | ebpf_write_end_chart(); |
| 1088 | } |
| 1089 | } |
| 1090 | |
| 1091 | /** |
| 1092 | * Create specific process charts |
| 1093 | * |
| 1094 | * Create charts for cgroup/application |
| 1095 | * |
| 1096 | * @param type the chart type. |
| 1097 | * @param em the structure with thread information |
| 1098 | */ |
| 1099 | static void ebpf_create_specific_process_charts(char *type, ebpf_module_t *em) |
| 1100 | { |
| 1101 | char *label = (!strncmp(type, "cgroup_", 7)) ? &type[7] : type; |
| 1102 | ebpf_create_chart( |
| 1103 | type, |
| 1104 | NETDATA_SYSCALL_APPS_TASK_PROCESS, |
| 1105 | "Process started", |
| 1106 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1107 | NETDATA_PROCESS_GROUP, |
| 1108 | NETDATA_CGROUP_PROCESS_CREATE_CONTEXT, |
| 1109 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1110 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5000, |
| 1111 | ebpf_create_global_dimension, |
| 1112 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_FORK], |
| 1113 | 1, |
| 1114 | em->update_every, |
| 1115 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 1116 | ebpf_create_chart_labels("cgroup_name", label, RRDLABEL_SRC_AUTO); |
| 1117 | ebpf_commit_label(); |
| 1118 | |
| 1119 | ebpf_create_chart( |
| 1120 | type, |
| 1121 | NETDATA_SYSCALL_APPS_TASK_THREAD, |
| 1122 | "Threads started", |
| 1123 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1124 | NETDATA_PROCESS_GROUP, |
| 1125 | NETDATA_CGROUP_THREAD_CREATE_CONTEXT, |
| 1126 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1127 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5001, |
| 1128 | ebpf_create_global_dimension, |
| 1129 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_CLONE], |
| 1130 | 1, |
| 1131 | em->update_every, |
| 1132 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 1133 | ebpf_create_chart_labels("cgroup_name", label, RRDLABEL_SRC_AUTO); |
| 1134 | ebpf_commit_label(); |
| 1135 | |
| 1136 | ebpf_create_chart( |
| 1137 | type, |
| 1138 | NETDATA_SYSCALL_APPS_TASK_EXIT, |
| 1139 | "Tasks starts exit process.", |
| 1140 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1141 | NETDATA_PROCESS_GROUP, |
| 1142 | NETDATA_CGROUP_PROCESS_EXIT_CONTEXT, |
| 1143 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1144 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5002, |
| 1145 | ebpf_create_global_dimension, |
| 1146 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT], |
| 1147 | 1, |
| 1148 | em->update_every, |
| 1149 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 1150 | ebpf_create_chart_labels("cgroup_name", label, RRDLABEL_SRC_AUTO); |
| 1151 | ebpf_commit_label(); |
| 1152 | |
| 1153 | ebpf_create_chart( |
| 1154 | type, |
| 1155 | NETDATA_SYSCALL_APPS_TASK_CLOSE, |
| 1156 | "Tasks closed", |
| 1157 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1158 | NETDATA_PROCESS_GROUP, |
| 1159 | NETDATA_CGROUP_PROCESS_CLOSE_CONTEXT, |
| 1160 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1161 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5003, |
| 1162 | ebpf_create_global_dimension, |
| 1163 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_RELEASE_TASK], |
| 1164 | 1, |
| 1165 | em->update_every, |
| 1166 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 1167 | ebpf_create_chart_labels("cgroup_name", label, RRDLABEL_SRC_AUTO); |
| 1168 | ebpf_commit_label(); |
| 1169 | |
| 1170 | if (em->mode < MODE_ENTRY) { |
| 1171 | ebpf_create_chart( |
| 1172 | type, |
| 1173 | NETDATA_SYSCALL_APPS_TASK_ERROR, |
| 1174 | "Errors to create process or threads.", |
| 1175 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1176 | NETDATA_PROCESS_GROUP, |
| 1177 | NETDATA_CGROUP_PROCESS_ERROR_CONTEXT, |
| 1178 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1179 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5004, |
| 1180 | ebpf_create_global_dimension, |
| 1181 | &process_publish_aggregated[NETDATA_KEY_PUBLISH_PROCESS_EXIT], |
| 1182 | 1, |
| 1183 | em->update_every, |
| 1184 | NETDATA_EBPF_MODULE_NAME_PROCESS); |
| 1185 | ebpf_create_chart_labels("cgroup_name", label, RRDLABEL_SRC_AUTO); |
| 1186 | ebpf_commit_label(); |
| 1187 | } |
| 1188 | } |
| 1189 | |
| 1190 | /** |
| 1191 | * Obsolete specific process charts |
| 1192 | * |
| 1193 | * Obsolete charts for cgroup/application |
| 1194 | * |
| 1195 | * @param type the chart type. |
| 1196 | * @param em the structure with thread information |
| 1197 | */ |
| 1198 | static void ebpf_obsolete_specific_process_charts(char *type, ebpf_module_t *em) |
| 1199 | { |
| 1200 | ebpf_write_chart_obsolete( |
| 1201 | type, |
| 1202 | NETDATA_SYSCALL_APPS_TASK_PROCESS, |
| 1203 | "", |
| 1204 | "Process started", |
| 1205 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1206 | NETDATA_PROCESS_GROUP, |
| 1207 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1208 | NETDATA_CGROUP_PROCESS_CREATE_CONTEXT, |
| 1209 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5000, |
| 1210 | em->update_every); |
| 1211 | |
| 1212 | ebpf_write_chart_obsolete( |
| 1213 | type, |
| 1214 | NETDATA_SYSCALL_APPS_TASK_THREAD, |
| 1215 | "", |
| 1216 | "Threads started", |
| 1217 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1218 | NETDATA_PROCESS_GROUP, |
| 1219 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1220 | NETDATA_CGROUP_THREAD_CREATE_CONTEXT, |
| 1221 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5001, |
| 1222 | em->update_every); |
| 1223 | |
| 1224 | ebpf_write_chart_obsolete( |
| 1225 | type, |
| 1226 | NETDATA_SYSCALL_APPS_TASK_EXIT, |
| 1227 | "", |
| 1228 | "Tasks starts exit process.", |
| 1229 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1230 | NETDATA_PROCESS_GROUP, |
| 1231 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1232 | NETDATA_CGROUP_PROCESS_EXIT_CONTEXT, |
| 1233 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5002, |
| 1234 | em->update_every); |
| 1235 | |
| 1236 | ebpf_write_chart_obsolete( |
| 1237 | type, |
| 1238 | NETDATA_SYSCALL_APPS_TASK_CLOSE, |
| 1239 | "", |
| 1240 | "Tasks closed", |
| 1241 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1242 | NETDATA_PROCESS_GROUP, |
| 1243 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1244 | NETDATA_CGROUP_PROCESS_CLOSE_CONTEXT, |
| 1245 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5003, |
| 1246 | em->update_every); |
| 1247 | |
| 1248 | if (em->mode < MODE_ENTRY) { |
| 1249 | ebpf_write_chart_obsolete( |
| 1250 | type, |
| 1251 | NETDATA_SYSCALL_APPS_TASK_ERROR, |
| 1252 | "", |
| 1253 | "Errors to create process or threads.", |
| 1254 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1255 | NETDATA_PROCESS_GROUP, |
| 1256 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1257 | NETDATA_CGROUP_PROCESS_ERROR_CONTEXT, |
| 1258 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5004, |
| 1259 | em->update_every); |
| 1260 | } |
| 1261 | } |
| 1262 | |
| 1263 | /** |
| 1264 | * Create Systemd process Charts |
| 1265 | * |
| 1266 | * Create charts when systemd is enabled |
| 1267 | * |
| 1268 | * @param em the structure with thread information |
| 1269 | **/ |
| 1270 | static void ebpf_create_systemd_process_charts(ebpf_module_t *em) |
| 1271 | { |
| 1272 | static ebpf_systemd_args_t data_process = { |
| 1273 | .title = "Process started", |
| 1274 | .units = EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1275 | .family = NETDATA_APPS_PROCESS_GROUP, |
| 1276 | .charttype = NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1277 | .order = 20065, |
| 1278 | .algorithm = EBPF_CHART_ALGORITHM_INCREMENTAL, |
| 1279 | .context = NETDATA_SYSTEMD_PROCESS_CREATE_CONTEXT, |
| 1280 | .module = NETDATA_EBPF_MODULE_NAME_PROCESS, |
| 1281 | .update_every = 0, |
| 1282 | .suffix = NETDATA_SYSCALL_APPS_TASK_PROCESS, |
| 1283 | .dimension = "calls"}; |
| 1284 | |
| 1285 | static ebpf_systemd_args_t data_thread = { |
| 1286 | .title = "Threads started", |
| 1287 | .units = EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1288 | .family = NETDATA_APPS_PROCESS_GROUP, |
| 1289 | .charttype = NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1290 | .order = 20066, |
| 1291 | .algorithm = EBPF_CHART_ALGORITHM_INCREMENTAL, |
| 1292 | .context = NETDATA_SYSTEMD_THREAD_CREATE_CONTEXT, |
| 1293 | .module = NETDATA_EBPF_MODULE_NAME_PROCESS, |
| 1294 | .update_every = 0, |
| 1295 | .suffix = NETDATA_SYSCALL_APPS_TASK_THREAD, |
| 1296 | .dimension = "calls"}; |
| 1297 | |
| 1298 | static ebpf_systemd_args_t task_exit = { |
| 1299 | .title = "Tasks starts exit process.", |
| 1300 | .units = EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1301 | .family = NETDATA_APPS_PROCESS_GROUP, |
| 1302 | .charttype = NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1303 | .order = 20067, |
| 1304 | .algorithm = EBPF_CHART_ALGORITHM_INCREMENTAL, |
| 1305 | .context = NETDATA_SYSTEMD_PROCESS_EXIT_CONTEXT, |
| 1306 | .module = NETDATA_EBPF_MODULE_NAME_PROCESS, |
| 1307 | .update_every = 0, |
| 1308 | .suffix = NETDATA_SYSCALL_APPS_TASK_CLOSE, |
| 1309 | .dimension = "calls"}; |
| 1310 | |
| 1311 | static ebpf_systemd_args_t task_closed = { |
| 1312 | .title = "Tasks closed", |
| 1313 | .units = EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1314 | .family = NETDATA_APPS_PROCESS_GROUP, |
| 1315 | .charttype = NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1316 | .order = 20068, |
| 1317 | .algorithm = EBPF_CHART_ALGORITHM_INCREMENTAL, |
| 1318 | .context = NETDATA_SYSTEMD_PROCESS_CLOSE_CONTEXT, |
| 1319 | .module = NETDATA_EBPF_MODULE_NAME_PROCESS, |
| 1320 | .update_every = 0, |
| 1321 | .suffix = NETDATA_SYSCALL_APPS_TASK_EXIT, |
| 1322 | .dimension = "calls"}; |
| 1323 | |
| 1324 | static ebpf_systemd_args_t task_error = { |
| 1325 | .title = "Errors to create process or threads.", |
| 1326 | .units = EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1327 | .family = NETDATA_APPS_PROCESS_GROUP, |
| 1328 | .charttype = NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1329 | .order = 20069, |
| 1330 | .algorithm = EBPF_CHART_ALGORITHM_INCREMENTAL, |
| 1331 | .context = NETDATA_SYSTEMD_PROCESS_ERROR_CONTEXT, |
| 1332 | .module = NETDATA_EBPF_MODULE_NAME_PROCESS, |
| 1333 | .update_every = 0, |
| 1334 | .suffix = NETDATA_SYSCALL_APPS_TASK_ERROR, |
| 1335 | .dimension = "calls"}; |
| 1336 | |
| 1337 | ebpf_cgroup_target_t *w; |
| 1338 | netdata_run_mode_t mode = em->mode; |
| 1339 | if (!task_exit.update_every) |
| 1340 | data_process.update_every = data_thread.update_every = task_exit.update_every = task_closed.update_every = |
| 1341 | task_error.update_every = em->update_every; |
| 1342 | |
| 1343 | for (w = ebpf_cgroup_pids; w; w = w->next) { |
| 1344 | if (ebpf_plugin_stop()) |
| 1345 | break; |
| 1346 | |
| 1347 | if (unlikely(!w->systemd || w->flags & NETDATA_EBPF_SERVICES_HAS_PROCESS_CHART)) |
| 1348 | continue; |
| 1349 | |
| 1350 | data_process.id = data_thread.id = task_exit.id = task_closed.id = task_error.id = w->name; |
| 1351 | ebpf_create_charts_on_systemd(&data_process); |
| 1352 | |
| 1353 | ebpf_create_charts_on_systemd(&data_thread); |
| 1354 | |
| 1355 | ebpf_create_charts_on_systemd(&task_exit); |
| 1356 | |
| 1357 | ebpf_create_charts_on_systemd(&task_closed); |
| 1358 | if (mode < MODE_ENTRY) { |
| 1359 | ebpf_create_charts_on_systemd(&task_error); |
| 1360 | } |
| 1361 | w->flags |= NETDATA_EBPF_SERVICES_HAS_PROCESS_CHART; |
| 1362 | } |
| 1363 | } |
| 1364 | |
| 1365 | /** |
| 1366 | * Send Systemd charts |
| 1367 | * |
| 1368 | * Send collected data to Netdata. |
| 1369 | * |
| 1370 | * @param em the structure with thread information |
| 1371 | */ |
| 1372 | static void ebpf_send_systemd_process_charts(ebpf_module_t *em) |
| 1373 | { |
| 1374 | ebpf_cgroup_target_t *ect; |
| 1375 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 1376 | if (ebpf_plugin_stop()) |
| 1377 | break; |
| 1378 | |
| 1379 | if (unlikely(!(ect->flags & NETDATA_EBPF_SERVICES_HAS_PROCESS_CHART))) { |
| 1380 | continue; |
| 1381 | } |
| 1382 | |
| 1383 | ebpf_write_begin_chart(ect->name, NETDATA_SYSCALL_APPS_TASK_PROCESS, ""); |
| 1384 | write_chart_dimension("calls", ect->publish_systemd_ps.create_process); |
| 1385 | ebpf_write_end_chart(); |
| 1386 | |
| 1387 | ebpf_write_begin_chart(ect->name, NETDATA_SYSCALL_APPS_TASK_THREAD, ""); |
| 1388 | write_chart_dimension("calls", ect->publish_systemd_ps.create_thread); |
| 1389 | ebpf_write_end_chart(); |
| 1390 | |
| 1391 | ebpf_write_begin_chart(ect->name, NETDATA_SYSCALL_APPS_TASK_EXIT, ""); |
| 1392 | write_chart_dimension("calls", ect->publish_systemd_ps.exit_call); |
| 1393 | ebpf_write_end_chart(); |
| 1394 | |
| 1395 | ebpf_write_begin_chart(ect->name, NETDATA_SYSCALL_APPS_TASK_CLOSE, ""); |
| 1396 | write_chart_dimension("calls", ect->publish_systemd_ps.release_call); |
| 1397 | ebpf_write_end_chart(); |
| 1398 | |
| 1399 | if (em->mode < MODE_ENTRY) { |
| 1400 | ebpf_write_begin_chart(ect->name, NETDATA_SYSCALL_APPS_TASK_ERROR, ""); |
| 1401 | write_chart_dimension("calls", ect->publish_systemd_ps.task_err); |
| 1402 | ebpf_write_end_chart(); |
| 1403 | } |
| 1404 | } |
| 1405 | } |
| 1406 | |
| 1407 | /** |
| 1408 | * Send data to Netdata calling auxiliary functions. |
| 1409 | * |
| 1410 | * @param em the structure with thread information |
| 1411 | */ |
| 1412 | static void ebpf_process_send_cgroup_data(ebpf_module_t *em) |
| 1413 | { |
| 1414 | netdata_mutex_lock(&mutex_cgroup_shm); |
| 1415 | ebpf_cgroup_target_t *ect; |
| 1416 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 1417 | ebpf_process_sum_cgroup_pids(&ect->publish_systemd_ps, ect->pids); |
| 1418 | } |
| 1419 | |
| 1420 | if (ebpf_plugin_stop()) { |
| 1421 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 1422 | return; |
| 1423 | } |
| 1424 | |
| 1425 | if (ebpf_cgroup_systemd_enabled_get()) { |
| 1426 | if (ebpf_send_cgroup_chart_get()) { |
| 1427 | ebpf_create_systemd_process_charts(em); |
| 1428 | } |
| 1429 | |
| 1430 | ebpf_send_systemd_process_charts(em); |
| 1431 | } |
| 1432 | |
| 1433 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 1434 | if (ebpf_plugin_stop()) |
| 1435 | break; |
| 1436 | |
| 1437 | if (ect->systemd) |
| 1438 | continue; |
| 1439 | |
| 1440 | if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_PROCESS_CHART) && ect->updated) { |
| 1441 | ebpf_create_specific_process_charts(ect->name, em); |
| 1442 | ect->flags |= NETDATA_EBPF_CGROUP_HAS_PROCESS_CHART; |
| 1443 | } |
| 1444 | |
| 1445 | if (ect->flags & NETDATA_EBPF_CGROUP_HAS_PROCESS_CHART) { |
| 1446 | if (ect->updated) { |
| 1447 | ebpf_send_specific_process_data(ect->name, &ect->publish_systemd_ps, em); |
| 1448 | } else { |
| 1449 | ebpf_obsolete_specific_process_charts(ect->name, em); |
| 1450 | ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_PROCESS_CHART; |
| 1451 | } |
| 1452 | } |
| 1453 | } |
| 1454 | |
| 1455 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 1456 | } |
| 1457 | |
| 1458 | /** |
| 1459 | * Update Cgroup algorithm |
| 1460 | * |
| 1461 | * Change algorithm from absolute to incremental |
| 1462 | */ |
| 1463 | void ebpf_process_update_cgroup_algorithm() |
| 1464 | { |
| 1465 | int i; |
| 1466 | for (i = 0; i < NETDATA_KEY_PUBLISH_PROCESS_END; i++) { |
| 1467 | netdata_publish_syscall_t *ptr = &process_publish_aggregated[i]; |
| 1468 | ptr->algorithm = ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]; |
| 1469 | } |
| 1470 | } |
| 1471 | |
| 1472 | /** |
| 1473 | * Process Accumulator |
| 1474 | * |
| 1475 | * Sum all values read from kernel and store in the first address. |
| 1476 | * |
| 1477 | * @param out the vector with read values. |
| 1478 | * @param maps_per_core do I need to read all cores? |
| 1479 | */ |
| 1480 | void ebpf_process_apps_accumulator(ebpf_process_stat_t *out, int maps_per_core) |
| 1481 | { |
| 1482 | int i, end = (maps_per_core) ? ebpf_nprocs : 1; |
| 1483 | ebpf_process_stat_t *total = &out[0]; |
| 1484 | uint64_t ct = total->ct; |
| 1485 | for (i = 1; i < end; i++) { |
| 1486 | if (ebpf_plugin_stop()) |
| 1487 | break; |
| 1488 | |
| 1489 | ebpf_process_stat_t *w = &out[i]; |
| 1490 | total->exit_call += w->exit_call; |
| 1491 | total->task_err += w->task_err; |
| 1492 | total->create_thread += w->create_thread; |
| 1493 | total->create_process += w->create_process; |
| 1494 | total->release_call += w->release_call; |
| 1495 | |
| 1496 | if (w->ct > ct) |
| 1497 | ct = w->ct; |
| 1498 | } |
| 1499 | total->ct = ct; |
| 1500 | } |
| 1501 | |
| 1502 | /** |
| 1503 | * Sum values for pid |
| 1504 | * |
| 1505 | * @param structure to store result. |
| 1506 | * @param root the structure with all available PIDs |
| 1507 | */ |
| 1508 | void ebpf_process_sum_values_for_pids(ebpf_process_stat_t *process, struct ebpf_pid_on_target *root) |
| 1509 | { |
| 1510 | memset(process, 0, sizeof(ebpf_process_stat_t)); |
| 1511 | for (; root; root = root->next) { |
| 1512 | if (ebpf_plugin_stop()) |
| 1513 | break; |
| 1514 | |
| 1515 | uint32_t pid = root->pid; |
| 1516 | netdata_ebpf_pid_stats_t *local_pid = netdata_ebpf_lookup_shm_pointer_unsafe(pid); |
| 1517 | if (!local_pid || !(local_pid->threads & (1U << (NETDATA_EBPF_PIDS_PROCESS_IDX << 1)))) |
| 1518 | continue; |
| 1519 | |
| 1520 | ebpf_publish_process_t *in = &local_pid->process; |
| 1521 | |
| 1522 | process->task_err += in->task_err; |
| 1523 | process->release_call += in->release_call; |
| 1524 | process->exit_call += in->exit_call; |
| 1525 | process->create_thread += in->create_thread; |
| 1526 | process->create_process += in->create_process; |
| 1527 | } |
| 1528 | } |
| 1529 | |
| 1530 | /** |
| 1531 | * Collect data for all process |
| 1532 | * |
| 1533 | * Read data from hash table and store it in appropriate vectors. |
| 1534 | * It also creates the link between targets and PIDs. |
| 1535 | * |
| 1536 | * @param tbl_pid_stats_fd The mapped file descriptor for the hash table. |
| 1537 | * @param maps_per_core do I have hash maps per core? |
| 1538 | */ |
| 1539 | void collect_data_for_all_processes(int tbl_pid_stats_fd, int maps_per_core) |
| 1540 | { |
| 1541 | if (tbl_pid_stats_fd == -1) |
| 1542 | return; |
| 1543 | |
| 1544 | ebpf_set_pid_map_fd(NETDATA_EBPF_PIDS_PROCESS_IDX, tbl_pid_stats_fd); |
| 1545 | size_t length = sizeof(ebpf_process_stat_t); |
| 1546 | if (maps_per_core) |
| 1547 | length *= ebpf_nprocs; |
| 1548 | |
| 1549 | if (tbl_pid_stats_fd != -1) { |
| 1550 | uint32_t key = 0, next_key = 0; |
| 1551 | while (bpf_map_get_next_key(tbl_pid_stats_fd, &key, &next_key) == 0) { |
| 1552 | if (ebpf_plugin_stop()) |
| 1553 | break; |
| 1554 | |
| 1555 | if (bpf_map_lookup_elem(tbl_pid_stats_fd, &key, process_stat_vector)) { |
| 1556 | goto end_process_loop; |
| 1557 | } |
| 1558 | |
| 1559 | ebpf_process_apps_accumulator(process_stat_vector, maps_per_core); |
| 1560 | |
| 1561 | netdata_ebpf_pid_stats_t *local_pid = |
| 1562 | netdata_ebpf_get_shm_pointer_unsafe(key, NETDATA_EBPF_PIDS_PROCESS_IDX); |
| 1563 | if (!local_pid) |
| 1564 | goto end_process_loop; |
| 1565 | |
| 1566 | ebpf_publish_process_t *w = &local_pid->process; |
| 1567 | |
| 1568 | if (!w->ct || w->ct != process_stat_vector[0].ct) { |
| 1569 | w->ct = process_stat_vector[0].ct; |
| 1570 | w->create_thread = process_stat_vector[0].create_thread; |
| 1571 | w->exit_call = process_stat_vector[0].exit_call; |
| 1572 | w->create_process = process_stat_vector[0].create_process; |
| 1573 | w->release_call = process_stat_vector[0].release_call; |
| 1574 | w->task_err = process_stat_vector[0].task_err; |
| 1575 | } else { |
| 1576 | if (kill((pid_t)key, 0) == -1 && errno == ESRCH) { |
| 1577 | if (netdata_ebpf_reset_shm_pointer_unsafe(tbl_pid_stats_fd, key, NETDATA_EBPF_PIDS_PROCESS_IDX)) |
| 1578 | memset(w, 0, sizeof(*w)); |
| 1579 | } |
| 1580 | } |
| 1581 | |
| 1582 | end_process_loop: |
| 1583 | memset(process_stat_vector, 0, length); |
| 1584 | key = next_key; |
| 1585 | } |
| 1586 | } |
| 1587 | |
| 1588 | struct ebpf_target *w; |
| 1589 | for (w = apps_groups_root_target; w; w = w->next) { |
| 1590 | if (ebpf_plugin_stop()) |
| 1591 | break; |
| 1592 | |
| 1593 | if (unlikely(!(w->processes))) |
| 1594 | continue; |
| 1595 | |
| 1596 | ebpf_process_sum_values_for_pids(&w->process, w->root_pid); |
| 1597 | } |
| 1598 | } |
| 1599 | |
| 1600 | /** |
| 1601 | * Main loop for this collector. |
| 1602 | * |
| 1603 | * @param em the structure with thread information |
| 1604 | */ |
| 1605 | static void process_collector(ebpf_module_t *em) |
| 1606 | { |
| 1607 | int publish_global = em->global_charts; |
| 1608 | int cgroups = em->cgroup_charts; |
| 1609 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 1610 | process_pid_fd = process_maps[NETDATA_PROCESS_PID_TABLE].map_fd; |
| 1611 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 1612 | if (cgroups) |
| 1613 | ebpf_process_update_cgroup_algorithm(); |
| 1614 | |
| 1615 | int update_every = em->update_every; |
| 1616 | int counter = update_every - 1; |
| 1617 | int maps_per_core = em->maps_per_core; |
| 1618 | uint32_t running_time = 0; |
| 1619 | uint32_t lifetime = em->lifetime; |
| 1620 | netdata_idx_t *stats = em->hash_table_stats; |
| 1621 | memset(stats, 0, sizeof(netdata_idx_t) * NETDATA_EBPF_GLOBAL_TABLE_STATUS_END); |
| 1622 | heartbeat_t hb; |
| 1623 | heartbeat_init(&hb, USEC_PER_SEC); |
| 1624 | int process_maps_per_core = ebpf_modules[EBPF_MODULE_PROCESS_IDX].maps_per_core; |
| 1625 | while (!ebpf_plugin_stop() && running_time < lifetime) { |
| 1626 | if (ebpf_plugin_stop()) |
| 1627 | break; |
| 1628 | |
| 1629 | heartbeat_next(&hb); |
| 1630 | |
| 1631 | if (ebpf_plugin_stop()) |
| 1632 | break; |
| 1633 | |
| 1634 | if (++counter == update_every) { |
| 1635 | counter = 0; |
| 1636 | |
| 1637 | ebpf_read_process_hash_global_tables(stats, maps_per_core); |
| 1638 | |
| 1639 | netdata_apps_integration_flags_t apps_enabled = em->apps_charts; |
| 1640 | |
| 1641 | if (ebpf_all_pids_count > 0) { |
| 1642 | if (!ebpf_shm_sem_wait_or_stop(shm_mutex_ebpf_integration)) { |
| 1643 | if (errno != ECANCELED) |
| 1644 | netdata_log_error("PROCESS: Failed to wait on semaphore."); |
| 1645 | break; |
| 1646 | } |
| 1647 | netdata_mutex_lock(&collect_data_mutex); |
| 1648 | collect_data_for_all_processes(process_pid_fd, process_maps_per_core); |
| 1649 | |
| 1650 | if (cgroups && ebpf_cgroup_integration_active_get()) { |
| 1651 | ebpf_update_process_cgroup(); |
| 1652 | } |
| 1653 | netdata_mutex_unlock(&collect_data_mutex); |
| 1654 | if (sem_post(shm_mutex_ebpf_integration)) { |
| 1655 | netdata_log_error("PROCESS: Failed to post semaphore."); |
| 1656 | break; |
| 1657 | } |
| 1658 | } |
| 1659 | |
| 1660 | // Avoid acquiring lock for data-send when shutdown is in progress; |
| 1661 | // the main thread may be holding lock for apps-parsing at this point, |
| 1662 | // which would delay this thread's join and compound the shutdown time. |
| 1663 | if (ebpf_plugin_stop()) |
| 1664 | break; |
| 1665 | |
| 1666 | netdata_mutex_lock(&lock); |
| 1667 | |
| 1668 | if (publish_global) { |
| 1669 | ebpf_process_send_data(em); |
| 1670 | } |
| 1671 | |
| 1672 | netdata_mutex_lock(&collect_data_mutex); |
| 1673 | if (apps_enabled & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) { |
| 1674 | ebpf_process_send_apps_data(apps_groups_root_target, em); |
| 1675 | } |
| 1676 | |
| 1677 | if (cgroups && ebpf_cgroup_integration_active_get()) { |
| 1678 | if (!ebpf_plugin_stop()) |
| 1679 | ebpf_process_send_cgroup_data(em); |
| 1680 | } |
| 1681 | |
| 1682 | netdata_mutex_unlock(&collect_data_mutex); |
| 1683 | netdata_mutex_unlock(&lock); |
| 1684 | |
| 1685 | fflush(stdout); |
| 1686 | |
| 1687 | if (ebpf_plugin_stop()) |
| 1688 | break; |
| 1689 | |
| 1690 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 1691 | running_time += update_every; |
| 1692 | em->running_time = running_time; |
| 1693 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 1694 | } |
| 1695 | } |
| 1696 | } |
| 1697 | |
| 1698 | /***************************************************************** |
| 1699 | * |
| 1700 | * FUNCTIONS TO START THREAD |
| 1701 | * |
| 1702 | *****************************************************************/ |
| 1703 | |
| 1704 | /** |
| 1705 | * Allocate vectors used with this thread. |
| 1706 | * We are not testing the return, because callocz does this and shutdown the software |
| 1707 | * case it was not possible to allocate. |
| 1708 | * |
| 1709 | * @param length is the length for the vectors used inside the collector. |
| 1710 | */ |
| 1711 | static void ebpf_process_allocate_global_vectors(size_t length) |
| 1712 | { |
| 1713 | memset(process_aggregated_data, 0, length * sizeof(netdata_syscall_stat_t)); |
| 1714 | memset(process_publish_aggregated, 0, length * sizeof(netdata_publish_syscall_t)); |
| 1715 | process_hash_values = callocz(ebpf_nprocs, sizeof(netdata_idx_t)); |
| 1716 | process_stat_vector = callocz(ebpf_nprocs, sizeof(ebpf_process_stat_t)); |
| 1717 | } |
| 1718 | |
| 1719 | static void change_syscalls() |
| 1720 | { |
| 1721 | static char *lfork = "do_fork"; |
| 1722 | process_id_names[NETDATA_KEY_PUBLISH_PROCESS_FORK] = lfork; |
| 1723 | } |
| 1724 | |
| 1725 | /** |
| 1726 | * Set local variables |
| 1727 | * |
| 1728 | */ |
| 1729 | static void set_local_pointers() |
| 1730 | { |
| 1731 | if (isrh >= NETDATA_MINIMUM_RH_VERSION && isrh < NETDATA_RH_8) |
| 1732 | change_syscalls(); |
| 1733 | } |
| 1734 | |
| 1735 | /***************************************************************** |
| 1736 | * |
| 1737 | * EBPF PROCESS THREAD |
| 1738 | * |
| 1739 | *****************************************************************/ |
| 1740 | |
| 1741 | /** |
| 1742 | * Enable a single tracepoint |
| 1743 | * |
| 1744 | * Enable a tracepoint and store whether it was already enabled. |
| 1745 | * |
| 1746 | * @param event the tracepoint event name |
| 1747 | * @param was_enabled pointer to store the previous state |
| 1748 | * |
| 1749 | * @return 0 on success, -1 on error |
| 1750 | */ |
| 1751 | static int ebpf_enable_single_tracepoint(const char *event, int *was_enabled) |
| 1752 | { |
| 1753 | int enabled = ebpf_is_tracepoint_enabled(tracepoint_sched_type, event); |
| 1754 | if (enabled == -1) |
| 1755 | return -1; |
| 1756 | if (!enabled) { |
| 1757 | if (ebpf_enable_tracing_values(tracepoint_sched_type, event)) |
| 1758 | return -1; |
| 1759 | } |
| 1760 | *was_enabled = enabled; |
| 1761 | return 0; |
| 1762 | } |
| 1763 | |
| 1764 | /** |
| 1765 | * Enable tracepoints |
| 1766 | * |
| 1767 | * Enable necessary tracepoints for thread. |
| 1768 | * |
| 1769 | * @return It returns 0 on success and -1 otherwise |
| 1770 | */ |
| 1771 | static int ebpf_process_enable_tracepoints() |
| 1772 | { |
| 1773 | if (ebpf_enable_single_tracepoint(tracepoint_sched_process_exit, &was_sched_process_exit_enabled)) |
| 1774 | return -1; |
| 1775 | |
| 1776 | if (ebpf_enable_single_tracepoint(tracepoint_sched_process_exec, &was_sched_process_exec_enabled)) |
| 1777 | return -1; |
| 1778 | |
| 1779 | if (ebpf_enable_single_tracepoint(tracepoint_sched_process_fork, &was_sched_process_fork_enabled)) |
| 1780 | return -1; |
| 1781 | |
| 1782 | return 0; |
| 1783 | } |
| 1784 | |
| 1785 | /** |
| 1786 | * Process thread |
| 1787 | * |
| 1788 | * Thread used to generate process charts. |
| 1789 | * |
| 1790 | * @param ptr a pointer to `struct ebpf_module` |
| 1791 | * |
| 1792 | * @return It always return NULL |
| 1793 | */ |
| 1794 | void ebpf_process_thread(void *ptr) |
| 1795 | { |
| 1796 | ebpf_module_t *em = (ebpf_module_t *)ptr; |
| 1797 | |
| 1798 | process_safe_clean = false; |
| 1799 | |
| 1800 | CLEANUP_FUNCTION_REGISTER(ebpf_process_exit) cleanup_ptr = em; |
| 1801 | |
| 1802 | if (!ebpf_module_thread_has_valid_state(em)) { |
| 1803 | em->global_charts = em->apps_charts = em->cgroup_charts = NETDATA_THREAD_EBPF_STOPPING; |
| 1804 | ebpf_module_enabled_set(em, NETDATA_THREAD_EBPF_STOPPING); |
| 1805 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 1806 | ebpf_update_disabled_plugin_stats(em); |
| 1807 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 1808 | return; |
| 1809 | } |
| 1810 | |
| 1811 | em->maps = process_maps; |
| 1812 | |
| 1813 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 1814 | if (ebpf_process_enable_tracepoints()) { |
| 1815 | em->global_charts = em->apps_charts = em->cgroup_charts = NETDATA_THREAD_EBPF_STOPPING; |
| 1816 | ebpf_module_enabled_set(em, NETDATA_THREAD_EBPF_STOPPING); |
| 1817 | } |
| 1818 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 1819 | |
| 1820 | ebpf_process_allocate_global_vectors(NETDATA_KEY_PUBLISH_PROCESS_END); |
| 1821 | netdata_mutex_lock(&lock); |
| 1822 | ebpf_update_pid_table(&process_maps[0], em); |
| 1823 | |
| 1824 | set_local_pointers(); |
| 1825 | if (ebpf_process_load_bpf(em)) { |
| 1826 | em->global_charts = em->apps_charts = em->cgroup_charts = NETDATA_THREAD_EBPF_STOPPING; |
| 1827 | ebpf_module_enabled_set(em, NETDATA_THREAD_EBPF_STOPPING); |
| 1828 | } else { |
| 1829 | ebpf_mark_program_loaded(); |
| 1830 | } |
| 1831 | |
| 1832 | int algorithms[NETDATA_KEY_PUBLISH_PROCESS_END] = { |
| 1833 | NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX, NETDATA_EBPF_ABSOLUTE_IDX}; |
| 1834 | |
| 1835 | ebpf_global_labels( |
| 1836 | process_aggregated_data, |
| 1837 | process_publish_aggregated, |
| 1838 | process_dimension_names, |
| 1839 | process_id_names, |
| 1840 | algorithms, |
| 1841 | NETDATA_KEY_PUBLISH_PROCESS_END); |
| 1842 | |
| 1843 | ebpf_create_global_charts(em); |
| 1844 | |
| 1845 | ebpf_update_stats(&plugin_statistics, em); |
| 1846 | ebpf_update_kernel_memory_with_vector(&plugin_statistics, em->maps, EBPF_ACTION_STAT_ADD); |
| 1847 | |
| 1848 | netdata_mutex_unlock(&lock); |
| 1849 | |
| 1850 | process_safe_clean = true; |
| 1851 | process_collector(em); |
| 1852 | |
| 1853 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 1854 | ebpf_update_disabled_plugin_stats(em); |
| 1855 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 1856 | } |