| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | #include "ebpf.h" |
| 4 | #include "ebpf_fd.h" |
| 5 | #include "libbpf_api/ebpf_library.h" |
| 6 | |
| 7 | static char *fd_dimension_names[NETDATA_FD_SYSCALL_END] = {"open", "close"}; |
| 8 | static char *fd_id_names[NETDATA_FD_SYSCALL_END] = {"do_sys_open", "__close_fd"}; |
| 9 | |
| 10 | #ifdef LIBBPF_MAJOR_VERSION |
| 11 | static char *close_targets[NETDATA_EBPF_MAX_FD_TARGETS] = {"close_fd", "__close_fd"}; |
| 12 | static char *open_targets[NETDATA_EBPF_MAX_FD_TARGETS] = {"do_sys_openat2", "do_sys_open"}; |
| 13 | #endif |
| 14 | |
| 15 | static netdata_syscall_stat_t fd_aggregated_data[NETDATA_FD_SYSCALL_END]; |
| 16 | static netdata_publish_syscall_t fd_publish_aggregated[NETDATA_FD_SYSCALL_END]; |
| 17 | |
| 18 | static ebpf_local_maps_t fd_maps[] = { |
| 19 | {.name = "tbl_fd_pid", |
| 20 | .internal_input = ND_EBPF_DEFAULT_PID_SIZE, |
| 21 | .user_input = 0, |
| 22 | .type = NETDATA_EBPF_MAP_RESIZABLE | NETDATA_EBPF_MAP_PID, |
| 23 | .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED, |
| 24 | #ifdef LIBBPF_MAJOR_VERSION |
| 25 | .map_type = BPF_MAP_TYPE_PERCPU_HASH |
| 26 | #endif |
| 27 | }, |
| 28 | {.name = "tbl_fd_global", |
| 29 | .internal_input = NETDATA_KEY_END_VECTOR, |
| 30 | .user_input = 0, |
| 31 | .type = NETDATA_EBPF_MAP_STATIC, |
| 32 | .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED, |
| 33 | #ifdef LIBBPF_MAJOR_VERSION |
| 34 | .map_type = BPF_MAP_TYPE_PERCPU_ARRAY |
| 35 | #endif |
| 36 | }, |
| 37 | {.name = "fd_ctrl", |
| 38 | .internal_input = NETDATA_CONTROLLER_END, |
| 39 | .user_input = 0, |
| 40 | .type = NETDATA_EBPF_MAP_CONTROLLER, |
| 41 | .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED, |
| 42 | #ifdef LIBBPF_MAJOR_VERSION |
| 43 | .map_type = BPF_MAP_TYPE_PERCPU_ARRAY |
| 44 | #endif |
| 45 | }, |
| 46 | {.name = NULL, |
| 47 | .internal_input = 0, |
| 48 | .user_input = 0, |
| 49 | .type = NETDATA_EBPF_MAP_CONTROLLER, |
| 50 | .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED, |
| 51 | #ifdef LIBBPF_MAJOR_VERSION |
| 52 | .map_type = BPF_MAP_TYPE_PERCPU_ARRAY |
| 53 | #endif |
| 54 | }}; |
| 55 | |
| 56 | struct config fd_config = APPCONFIG_INITIALIZER; |
| 57 | |
| 58 | static netdata_idx_t fd_hash_values[NETDATA_FD_COUNTER]; |
| 59 | static netdata_idx_t *fd_values = NULL; |
| 60 | |
| 61 | netdata_fd_stat_t *fd_vector = NULL; |
| 62 | static bool fd_safe_clean = false; |
| 63 | |
| 64 | static int fd_use_close_fd = -1; |
| 65 | |
| 66 | netdata_ebpf_targets_t fd_targets[] = { |
| 67 | {.name = "open", .mode = EBPF_LOAD_TRAMPOLINE}, |
| 68 | {.name = "close", .mode = EBPF_LOAD_TRAMPOLINE}, |
| 69 | {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; |
| 70 | |
| 71 | struct netdata_static_thread ebpf_read_fd = { |
| 72 | .name = "EBPF_READ_FD", |
| 73 | .config_section = NULL, |
| 74 | .config_name = NULL, |
| 75 | .env_name = NULL, |
| 76 | .enabled = 1, |
| 77 | .thread = NULL, |
| 78 | .init_routine = NULL, |
| 79 | .start_routine = NULL}; |
| 80 | |
| 81 | #ifdef LIBBPF_MAJOR_VERSION |
| 82 | /** |
| 83 | * Check if using close_fd |
| 84 | * |
| 85 | * @return true if using close_fd, false if using __close_fd |
| 86 | */ |
| 87 | static inline int ebpf_fd_using_close_fd(void) |
| 88 | { |
| 89 | if (fd_use_close_fd == -1) |
| 90 | fd_use_close_fd = !strcmp(fd_targets[NETDATA_FD_SYSCALL_CLOSE].name, close_targets[NETDATA_FD_CLOSE_FD]); |
| 91 | return fd_use_close_fd; |
| 92 | } |
| 93 | |
| 94 | /** |
| 95 | * Disable probe |
| 96 | * |
| 97 | * Disable all probes to use exclusively another method. |
| 98 | * |
| 99 | * @param obj is the main structure for bpf objects |
| 100 | */ |
| 101 | static inline void ebpf_fd_disable_probes(struct fd_bpf *obj) |
| 102 | { |
| 103 | bpf_program__set_autoload(obj->progs.netdata_sys_open_kprobe, false); |
| 104 | bpf_program__set_autoload(obj->progs.netdata_sys_open_kretprobe, false); |
| 105 | if (ebpf_fd_using_close_fd()) { |
| 106 | bpf_program__set_autoload(obj->progs.netdata___close_fd_kretprobe, false); |
| 107 | bpf_program__set_autoload(obj->progs.netdata___close_fd_kprobe, false); |
| 108 | bpf_program__set_autoload(obj->progs.netdata_close_fd_kprobe, false); |
| 109 | } else { |
| 110 | bpf_program__set_autoload(obj->progs.netdata___close_fd_kprobe, false); |
| 111 | bpf_program__set_autoload(obj->progs.netdata_close_fd_kretprobe, false); |
| 112 | bpf_program__set_autoload(obj->progs.netdata_close_fd_kprobe, false); |
| 113 | } |
| 114 | } |
| 115 | |
| 116 | /* |
| 117 | * Disable specific probe |
| 118 | * |
| 119 | * Disable probes according the kernel version |
| 120 | * |
| 121 | * @param obj is the main structure for bpf objects |
| 122 | */ |
| 123 | static inline void ebpf_disable_specific_probes(struct fd_bpf *obj) |
| 124 | { |
| 125 | if (ebpf_fd_using_close_fd()) { |
| 126 | bpf_program__set_autoload(obj->progs.netdata___close_fd_kretprobe, false); |
| 127 | bpf_program__set_autoload(obj->progs.netdata___close_fd_kprobe, false); |
| 128 | } else { |
| 129 | bpf_program__set_autoload(obj->progs.netdata_close_fd_kretprobe, false); |
| 130 | bpf_program__set_autoload(obj->progs.netdata_close_fd_kprobe, false); |
| 131 | } |
| 132 | } |
| 133 | |
| 134 | /* |
| 135 | * Disable trampoline |
| 136 | * |
| 137 | * Disable all trampoline to use exclusively another method. |
| 138 | * |
| 139 | * @param obj is the main structure for bpf objects. |
| 140 | */ |
| 141 | static inline void ebpf_disable_trampoline(struct fd_bpf *obj) |
| 142 | { |
| 143 | bpf_program__set_autoload(obj->progs.netdata_sys_open_fentry, false); |
| 144 | bpf_program__set_autoload(obj->progs.netdata_sys_open_fexit, false); |
| 145 | bpf_program__set_autoload(obj->progs.netdata_close_fd_fentry, false); |
| 146 | bpf_program__set_autoload(obj->progs.netdata_close_fd_fexit, false); |
| 147 | bpf_program__set_autoload(obj->progs.netdata___close_fd_fentry, false); |
| 148 | bpf_program__set_autoload(obj->progs.netdata___close_fd_fexit, false); |
| 149 | } |
| 150 | |
| 151 | /* |
| 152 | * Disable specific trampoline |
| 153 | * |
| 154 | * Disable trampoline according to kernel version. |
| 155 | * |
| 156 | * @param obj is the main structure for bpf objects. |
| 157 | */ |
| 158 | static inline void ebpf_disable_specific_trampoline(struct fd_bpf *obj) |
| 159 | { |
| 160 | if (ebpf_fd_using_close_fd()) { |
| 161 | bpf_program__set_autoload(obj->progs.netdata___close_fd_fentry, false); |
| 162 | bpf_program__set_autoload(obj->progs.netdata___close_fd_fexit, false); |
| 163 | } else { |
| 164 | bpf_program__set_autoload(obj->progs.netdata_close_fd_fentry, false); |
| 165 | bpf_program__set_autoload(obj->progs.netdata_close_fd_fexit, false); |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | /** |
| 170 | * Set trampoline target |
| 171 | * |
| 172 | * Set the targets we will monitor. |
| 173 | * |
| 174 | * @param obj is the main structure for bpf objects. |
| 175 | */ |
| 176 | static void ebpf_set_trampoline_target(struct fd_bpf *obj) |
| 177 | { |
| 178 | bpf_program__set_attach_target(obj->progs.netdata_sys_open_fentry, 0, fd_targets[NETDATA_FD_SYSCALL_OPEN].name); |
| 179 | bpf_program__set_attach_target(obj->progs.netdata_sys_open_fexit, 0, fd_targets[NETDATA_FD_SYSCALL_OPEN].name); |
| 180 | |
| 181 | if (ebpf_fd_using_close_fd()) { |
| 182 | bpf_program__set_attach_target( |
| 183 | obj->progs.netdata_close_fd_fentry, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); |
| 184 | bpf_program__set_attach_target(obj->progs.netdata_close_fd_fexit, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); |
| 185 | } else { |
| 186 | bpf_program__set_attach_target( |
| 187 | obj->progs.netdata___close_fd_fentry, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); |
| 188 | bpf_program__set_attach_target( |
| 189 | obj->progs.netdata___close_fd_fexit, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); |
| 190 | } |
| 191 | } |
| 192 | |
| 193 | /** |
| 194 | * Mount Attach Probe |
| 195 | * |
| 196 | * Attach probes to target |
| 197 | * |
| 198 | * @param obj is the main structure for bpf objects. |
| 199 | * |
| 200 | * @return It returns 0 on success and -1 otherwise. |
| 201 | */ |
| 202 | static int ebpf_fd_attach_probe(struct fd_bpf *obj) |
| 203 | { |
| 204 | obj->links.netdata_sys_open_kprobe = |
| 205 | bpf_program__attach_kprobe(obj->progs.netdata_sys_open_kprobe, false, fd_targets[NETDATA_FD_SYSCALL_OPEN].name); |
| 206 | long ret = libbpf_get_error(obj->links.netdata_sys_open_kprobe); |
| 207 | if (ret) |
| 208 | return -1; |
| 209 | |
| 210 | obj->links.netdata_sys_open_kretprobe = bpf_program__attach_kprobe( |
| 211 | obj->progs.netdata_sys_open_kretprobe, true, fd_targets[NETDATA_FD_SYSCALL_OPEN].name); |
| 212 | ret = libbpf_get_error(obj->links.netdata_sys_open_kretprobe); |
| 213 | if (ret) |
| 214 | return -1; |
| 215 | |
| 216 | if (ebpf_fd_using_close_fd()) { |
| 217 | obj->links.netdata_close_fd_kretprobe = bpf_program__attach_kprobe( |
| 218 | obj->progs.netdata_close_fd_kretprobe, true, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); |
| 219 | ret = libbpf_get_error(obj->links.netdata_close_fd_kretprobe); |
| 220 | if (ret) |
| 221 | return -1; |
| 222 | |
| 223 | obj->links.netdata_close_fd_kprobe = bpf_program__attach_kprobe( |
| 224 | obj->progs.netdata_close_fd_kprobe, false, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); |
| 225 | ret = libbpf_get_error(obj->links.netdata_close_fd_kprobe); |
| 226 | if (ret) |
| 227 | return -1; |
| 228 | } else { |
| 229 | obj->links.netdata___close_fd_kretprobe = bpf_program__attach_kprobe( |
| 230 | obj->progs.netdata___close_fd_kretprobe, true, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); |
| 231 | ret = libbpf_get_error(obj->links.netdata___close_fd_kretprobe); |
| 232 | if (ret) |
| 233 | return -1; |
| 234 | |
| 235 | obj->links.netdata___close_fd_kprobe = bpf_program__attach_kprobe( |
| 236 | obj->progs.netdata___close_fd_kprobe, false, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); |
| 237 | ret = libbpf_get_error(obj->links.netdata___close_fd_kprobe); |
| 238 | if (ret) |
| 239 | return -1; |
| 240 | } |
| 241 | |
| 242 | return 0; |
| 243 | } |
| 244 | |
| 245 | /** |
| 246 | * FD Fill Address |
| 247 | * |
| 248 | * Fill address value used to load probes/trampoline. |
| 249 | */ |
| 250 | static inline void ebpf_fd_fill_address(ebpf_addresses_t *address, char **targets) |
| 251 | { |
| 252 | int i; |
| 253 | for (i = 0; i < NETDATA_EBPF_MAX_FD_TARGETS; i++) { |
| 254 | address->function = targets[i]; |
| 255 | ebpf_load_addresses(address, -1); |
| 256 | if (address->addr) |
| 257 | break; |
| 258 | } |
| 259 | } |
| 260 | |
| 261 | /** |
| 262 | * Set target values |
| 263 | * |
| 264 | * Set pointers used to load data. |
| 265 | * |
| 266 | * @return It returns 0 on success and -1 otherwise. |
| 267 | */ |
| 268 | static int ebpf_fd_set_target_values(void) |
| 269 | { |
| 270 | ebpf_addresses_t address = {.function = NULL, .hash = 0, .addr = 0}; |
| 271 | ebpf_fd_fill_address(&address, close_targets); |
| 272 | |
| 273 | if (!address.addr) |
| 274 | return -1; |
| 275 | |
| 276 | fd_targets[NETDATA_FD_SYSCALL_CLOSE].name = address.function; |
| 277 | |
| 278 | address.addr = 0; |
| 279 | ebpf_fd_fill_address(&address, open_targets); |
| 280 | |
| 281 | if (!address.addr) |
| 282 | return -1; |
| 283 | |
| 284 | fd_targets[NETDATA_FD_SYSCALL_OPEN].name = address.function; |
| 285 | |
| 286 | return 0; |
| 287 | } |
| 288 | |
| 289 | /** |
| 290 | * Set hash tables |
| 291 | * |
| 292 | * Set the values for maps according the value given by kernel. |
| 293 | * |
| 294 | * @param obj is the main structure for bpf objects. |
| 295 | */ |
| 296 | static int ebpf_fd_set_hash_tables(struct fd_bpf *obj) |
| 297 | { |
| 298 | int map_fd; |
| 299 | |
| 300 | map_fd = bpf_map__fd(obj->maps.tbl_fd_global); |
| 301 | if (map_fd < 0) { |
| 302 | netdata_log_error("Failed to get fd for tbl_fd_global map"); |
| 303 | return -1; |
| 304 | } |
| 305 | fd_maps[NETDATA_FD_GLOBAL_STATS].map_fd = map_fd; |
| 306 | |
| 307 | map_fd = bpf_map__fd(obj->maps.tbl_fd_pid); |
| 308 | if (map_fd < 0) { |
| 309 | netdata_log_error("Failed to get fd for tbl_fd_pid map"); |
| 310 | return -1; |
| 311 | } |
| 312 | fd_maps[NETDATA_FD_PID_STATS].map_fd = map_fd; |
| 313 | |
| 314 | map_fd = bpf_map__fd(obj->maps.fd_ctrl); |
| 315 | if (map_fd < 0) { |
| 316 | netdata_log_error("Failed to get fd for fd_ctrl map"); |
| 317 | return -1; |
| 318 | } |
| 319 | fd_maps[NETDATA_FD_CONTROLLER].map_fd = map_fd; |
| 320 | |
| 321 | return 0; |
| 322 | } |
| 323 | |
| 324 | /** |
| 325 | * Adjust Map Size |
| 326 | * |
| 327 | * Resize maps according input from users. |
| 328 | * |
| 329 | * @param obj is the main structure for bpf objects. |
| 330 | * @param em structure with configuration |
| 331 | */ |
| 332 | static void ebpf_fd_adjust_map(struct fd_bpf *obj, ebpf_module_t *em) |
| 333 | { |
| 334 | ebpf_update_map_size(obj->maps.tbl_fd_pid, &fd_maps[NETDATA_FD_PID_STATS], em, bpf_map__name(obj->maps.tbl_fd_pid)); |
| 335 | |
| 336 | ebpf_update_map_type(obj->maps.tbl_fd_global, &fd_maps[NETDATA_FD_GLOBAL_STATS]); |
| 337 | ebpf_update_map_type(obj->maps.tbl_fd_pid, &fd_maps[NETDATA_FD_PID_STATS]); |
| 338 | ebpf_update_map_type(obj->maps.fd_ctrl, &fd_maps[NETDATA_FD_CONTROLLER]); |
| 339 | } |
| 340 | |
| 341 | /** |
| 342 | * Load and attach |
| 343 | * |
| 344 | * Load and attach the eBPF code in kernel. |
| 345 | * |
| 346 | * @param obj is the main structure for bpf objects. |
| 347 | * @param em structure with configuration |
| 348 | * |
| 349 | * @return it returns 0 on success and -1 otherwise |
| 350 | */ |
| 351 | static inline int ebpf_fd_load_and_attach(struct fd_bpf *obj, ebpf_module_t *em) |
| 352 | { |
| 353 | netdata_ebpf_targets_t *mt = em->targets; |
| 354 | netdata_ebpf_program_loaded_t test = mt[NETDATA_FD_SYSCALL_OPEN].mode; |
| 355 | |
| 356 | if (ebpf_fd_set_target_values()) { |
| 357 | netdata_log_error("%s file descriptor.", NETDATA_EBPF_DEFAULT_FNT_NOT_FOUND); |
| 358 | return -1; |
| 359 | } |
| 360 | |
| 361 | if (test == EBPF_LOAD_TRAMPOLINE) { |
| 362 | ebpf_fd_disable_probes(obj); |
| 363 | ebpf_disable_specific_trampoline(obj); |
| 364 | |
| 365 | ebpf_set_trampoline_target(obj); |
| 366 | } else { |
| 367 | ebpf_disable_trampoline(obj); |
| 368 | ebpf_disable_specific_probes(obj); |
| 369 | } |
| 370 | |
| 371 | ebpf_fd_adjust_map(obj, em); |
| 372 | |
| 373 | int ret = fd_bpf__load(obj); |
| 374 | if (ret) { |
| 375 | return ret; |
| 376 | } |
| 377 | |
| 378 | ret = (test == EBPF_LOAD_TRAMPOLINE) ? fd_bpf__attach(obj) : ebpf_fd_attach_probe(obj); |
| 379 | if (!ret) { |
| 380 | ret = ebpf_fd_set_hash_tables(obj); |
| 381 | |
| 382 | if (!ret) { |
| 383 | ebpf_update_controller(fd_maps[NETDATA_FD_CONTROLLER].map_fd, em); |
| 384 | } |
| 385 | } |
| 386 | |
| 387 | return ret; |
| 388 | } |
| 389 | #endif |
| 390 | |
| 391 | /***************************************************************** |
| 392 | * |
| 393 | * FUNCTIONS TO CLOSE THE THREAD |
| 394 | * |
| 395 | *****************************************************************/ |
| 396 | |
| 397 | static void ebpf_obsolete_specific_fd_charts(char *type, ebpf_module_t *em); |
| 398 | |
| 399 | /** |
| 400 | * Obsolete services |
| 401 | * |
| 402 | * Obsolete all service charts created |
| 403 | * |
| 404 | * @param em a pointer to `struct ebpf_module` |
| 405 | */ |
| 406 | static void ebpf_obsolete_fd_services(ebpf_module_t *em, char *id) |
| 407 | { |
| 408 | ebpf_write_chart_obsolete( |
| 409 | id, |
| 410 | NETDATA_SYSCALL_APPS_FILE_OPEN, |
| 411 | "", |
| 412 | "Number of open files", |
| 413 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 414 | NETDATA_APPS_FILE_GROUP, |
| 415 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 416 | NETDATA_CGROUP_FD_OPEN_CONTEXT, |
| 417 | NETDATA_EBPF_FD_CHART_PRIORITY_OPEN, |
| 418 | em->update_every); |
| 419 | |
| 420 | if (em->mode < MODE_ENTRY) { |
| 421 | ebpf_write_chart_obsolete( |
| 422 | id, |
| 423 | NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR, |
| 424 | "", |
| 425 | "Fails to open files", |
| 426 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 427 | NETDATA_APPS_FILE_GROUP, |
| 428 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 429 | NETDATA_CGROUP_FD_OPEN_ERR_CONTEXT, |
| 430 | NETDATA_EBPF_FD_CHART_PRIORITY_OPEN_ERR, |
| 431 | em->update_every); |
| 432 | } |
| 433 | |
| 434 | ebpf_write_chart_obsolete( |
| 435 | id, |
| 436 | NETDATA_SYSCALL_APPS_FILE_CLOSED, |
| 437 | "", |
| 438 | "Files closed", |
| 439 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 440 | NETDATA_APPS_FILE_GROUP, |
| 441 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 442 | NETDATA_CGROUP_FD_CLOSE_CONTEXT, |
| 443 | NETDATA_EBPF_FD_CHART_PRIORITY_CLOSE, |
| 444 | em->update_every); |
| 445 | |
| 446 | if (em->mode < MODE_ENTRY) { |
| 447 | ebpf_write_chart_obsolete( |
| 448 | id, |
| 449 | NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR, |
| 450 | "", |
| 451 | "Fails to close files", |
| 452 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 453 | NETDATA_APPS_FILE_GROUP, |
| 454 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 455 | NETDATA_CGROUP_FD_CLOSE_ERR_CONTEXT, |
| 456 | NETDATA_EBPF_FD_CHART_PRIORITY_CLOSE_ERR, |
| 457 | em->update_every); |
| 458 | } |
| 459 | } |
| 460 | |
| 461 | /** |
| 462 | * Obsolete cgroup chart |
| 463 | * |
| 464 | * Send obsolete for all charts created before to close. |
| 465 | * |
| 466 | * @param em a pointer to `struct ebpf_module` |
| 467 | */ |
| 468 | static inline void ebpf_obsolete_fd_cgroup_charts(ebpf_module_t *em) |
| 469 | { |
| 470 | netdata_mutex_lock(&mutex_cgroup_shm); |
| 471 | |
| 472 | ebpf_cgroup_target_t *ect; |
| 473 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 474 | if (ect->systemd) { |
| 475 | ebpf_obsolete_fd_services(em, ect->name); |
| 476 | |
| 477 | continue; |
| 478 | } |
| 479 | |
| 480 | ebpf_obsolete_specific_fd_charts(ect->name, em); |
| 481 | } |
| 482 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 483 | } |
| 484 | |
| 485 | /** |
| 486 | * Obsolette apps charts |
| 487 | * |
| 488 | * Obsolete apps charts. |
| 489 | * |
| 490 | * @param em a pointer to the structure with the default values. |
| 491 | */ |
| 492 | void ebpf_obsolete_fd_apps_charts(struct ebpf_module *em) |
| 493 | { |
| 494 | struct ebpf_target *w; |
| 495 | int update_every = em->update_every; |
| 496 | netdata_mutex_lock(&collect_data_mutex); |
| 497 | for (w = apps_groups_root_target; w; w = w->next) { |
| 498 | if (unlikely(!(w->charts_created & (1 << EBPF_MODULE_FD_IDX)))) |
| 499 | continue; |
| 500 | |
| 501 | ebpf_write_chart_obsolete( |
| 502 | NETDATA_APP_FAMILY, |
| 503 | w->clean_name, |
| 504 | "_ebpf_file_open", |
| 505 | "Number of open files", |
| 506 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 507 | NETDATA_APPS_FILE_FDS, |
| 508 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 509 | "app.ebpf_file_open", |
| 510 | NETDATA_EBPF_FD_CHART_PRIORITY_APPS_OPEN, |
| 511 | update_every); |
| 512 | |
| 513 | if (em->mode < MODE_ENTRY) { |
| 514 | ebpf_write_chart_obsolete( |
| 515 | NETDATA_APP_FAMILY, |
| 516 | w->clean_name, |
| 517 | "_ebpf_file_open_error", |
| 518 | "Fails to open files.", |
| 519 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 520 | NETDATA_APPS_FILE_FDS, |
| 521 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 522 | "app.ebpf_file_open_error", |
| 523 | NETDATA_EBPF_FD_CHART_PRIORITY_APPS_OPEN_ERR, |
| 524 | update_every); |
| 525 | } |
| 526 | |
| 527 | ebpf_write_chart_obsolete( |
| 528 | NETDATA_APPS_FAMILY, |
| 529 | w->clean_name, |
| 530 | "_ebpf_file_closed", |
| 531 | "Files closed.", |
| 532 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 533 | NETDATA_APPS_FILE_FDS, |
| 534 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 535 | "app.ebpf_file_closed", |
| 536 | NETDATA_EBPF_FD_CHART_PRIORITY_APPS_CLOSE, |
| 537 | update_every); |
| 538 | |
| 539 | if (em->mode < MODE_ENTRY) { |
| 540 | ebpf_write_chart_obsolete( |
| 541 | NETDATA_APPS_FAMILY, |
| 542 | w->clean_name, |
| 543 | "_ebpf_file_close_error", |
| 544 | "Fails to close files.", |
| 545 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 546 | NETDATA_APPS_FILE_FDS, |
| 547 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 548 | "app.ebpf_fd_close_error", |
| 549 | NETDATA_EBPF_FD_CHART_PRIORITY_APPS_CLOSE_ERR, |
| 550 | update_every); |
| 551 | } |
| 552 | w->charts_created &= ~(1 << EBPF_MODULE_FD_IDX); |
| 553 | } |
| 554 | netdata_mutex_unlock(&collect_data_mutex); |
| 555 | } |
| 556 | |
| 557 | /** |
| 558 | * Obsolete global |
| 559 | * |
| 560 | * Obsolete global charts created by thread. |
| 561 | * |
| 562 | * @param em a pointer to `struct ebpf_module` |
| 563 | */ |
| 564 | static void ebpf_obsolete_fd_global(ebpf_module_t *em) |
| 565 | { |
| 566 | ebpf_write_chart_obsolete( |
| 567 | NETDATA_FILESYSTEM_FAMILY, |
| 568 | NETDATA_FILE_OPEN_CLOSE_COUNT, |
| 569 | "", |
| 570 | "Open and close calls", |
| 571 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 572 | NETDATA_FILE_GROUP, |
| 573 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 574 | NETDATA_FS_FILEDESCRIPTOR_CONTEXT, |
| 575 | NETDATA_CHART_PRIO_EBPF_FD_CHARTS, |
| 576 | em->update_every); |
| 577 | |
| 578 | if (em->mode < MODE_ENTRY) { |
| 579 | ebpf_write_chart_obsolete( |
| 580 | NETDATA_FILESYSTEM_FAMILY, |
| 581 | NETDATA_FILE_OPEN_ERR_COUNT, |
| 582 | "", |
| 583 | "Open fails", |
| 584 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 585 | NETDATA_FILE_GROUP, |
| 586 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 587 | NETDATA_FS_FILEDESCRIPTOR_ERROR_CONTEXT, |
| 588 | NETDATA_CHART_PRIO_EBPF_FD_CHARTS + 1, |
| 589 | em->update_every); |
| 590 | } |
| 591 | } |
| 592 | |
| 593 | /** |
| 594 | * FD Exit |
| 595 | * |
| 596 | * Cancel child thread and exit. |
| 597 | * |
| 598 | * @param ptr thread data. |
| 599 | */ |
| 600 | void ebpf_fd_unload_bpf(ebpf_module_t *em) |
| 601 | { |
| 602 | #ifdef LIBBPF_MAJOR_VERSION |
| 603 | if (fd_bpf_obj) { |
| 604 | fd_bpf__destroy(fd_bpf_obj); |
| 605 | fd_bpf_obj = NULL; |
| 606 | } |
| 607 | #endif |
| 608 | if ((em->load & EBPF_LOAD_LEGACY) && em->probe_links) { |
| 609 | ebpf_unload_legacy_code(em->objects, em->probe_links); |
| 610 | em->objects = NULL; |
| 611 | em->probe_links = NULL; |
| 612 | } |
| 613 | } |
| 614 | |
| 615 | static void ebpf_fd_exit(void *pptr) |
| 616 | { |
| 617 | ebpf_set_pid_map_fd(NETDATA_EBPF_PIDS_FD_IDX, -1); |
| 618 | ebpf_module_t *em = CLEANUP_FUNCTION_GET_PTR(pptr); |
| 619 | if (!em) |
| 620 | return; |
| 621 | |
| 622 | if (!fd_safe_clean) { |
| 623 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 624 | ebpf_module_enabled_set(em, NETDATA_THREAD_EBPF_STOPPED); |
| 625 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 626 | return; |
| 627 | } |
| 628 | |
| 629 | netdata_mutex_lock(&lock); |
| 630 | collect_pids &= ~(1 << EBPF_MODULE_FD_IDX); |
| 631 | netdata_mutex_unlock(&lock); |
| 632 | |
| 633 | if (ebpf_read_fd.thread) { |
| 634 | nd_thread_signal_cancel(ebpf_read_fd.thread); |
| 635 | nd_thread_join(ebpf_read_fd.thread); |
| 636 | } |
| 637 | |
| 638 | // Drop this module's bits from the shared PID pool so its slots don't |
| 639 | // stay pinned if the plugin keeps running after the module stops. |
| 640 | if (integration_shm && ebpf_shm_sem_wait_or_stop(shm_mutex_ebpf_integration)) { |
| 641 | netdata_ebpf_sweep_shm_for_module_unsafe(NETDATA_EBPF_PIDS_FD_IDX); |
| 642 | sem_post(shm_mutex_ebpf_integration); |
| 643 | } |
| 644 | |
| 645 | if (ebpf_module_enabled_get(em) == NETDATA_THREAD_EBPF_FUNCTION_RUNNING && !ebpf_plugin_stop()) { |
| 646 | netdata_mutex_lock(&lock); |
| 647 | if (em->cgroup_charts) { |
| 648 | ebpf_obsolete_fd_cgroup_charts(em); |
| 649 | fflush(stdout); |
| 650 | } |
| 651 | |
| 652 | if (em->apps_charts & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) { |
| 653 | ebpf_obsolete_fd_apps_charts(em); |
| 654 | } |
| 655 | |
| 656 | ebpf_obsolete_fd_global(em); |
| 657 | |
| 658 | fflush(stdout); |
| 659 | netdata_mutex_unlock(&lock); |
| 660 | } |
| 661 | |
| 662 | freez(fd_vector); |
| 663 | fd_vector = NULL; |
| 664 | freez(fd_values); |
| 665 | fd_values = NULL; |
| 666 | |
| 667 | if (!ebpf_plugin_stop() && em->functions.bpf_unload) |
| 668 | em->functions.bpf_unload(em); |
| 669 | |
| 670 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 671 | ebpf_module_enabled_set(em, NETDATA_THREAD_EBPF_STOPPED); |
| 672 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 673 | } |
| 674 | |
| 675 | /***************************************************************** |
| 676 | * |
| 677 | * MAIN LOOP |
| 678 | * |
| 679 | *****************************************************************/ |
| 680 | |
| 681 | /** |
| 682 | * Send data to Netdata calling auxiliary functions. |
| 683 | * |
| 684 | * @param em the structure with thread information |
| 685 | */ |
| 686 | static void ebpf_fd_send_data(ebpf_module_t *em) |
| 687 | { |
| 688 | fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN].ncall = fd_hash_values[NETDATA_KEY_CALLS_DO_SYS_OPEN]; |
| 689 | fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN].nerr = fd_hash_values[NETDATA_KEY_ERROR_DO_SYS_OPEN]; |
| 690 | |
| 691 | fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE].ncall = fd_hash_values[NETDATA_KEY_CALLS_CLOSE_FD]; |
| 692 | fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE].nerr = fd_hash_values[NETDATA_KEY_ERROR_CLOSE_FD]; |
| 693 | |
| 694 | write_count_chart( |
| 695 | NETDATA_FILE_OPEN_CLOSE_COUNT, NETDATA_FILESYSTEM_FAMILY, fd_publish_aggregated, NETDATA_FD_SYSCALL_END); |
| 696 | |
| 697 | if (em->mode < MODE_ENTRY) { |
| 698 | write_err_chart( |
| 699 | NETDATA_FILE_OPEN_ERR_COUNT, NETDATA_FILESYSTEM_FAMILY, fd_publish_aggregated, NETDATA_FD_SYSCALL_END); |
| 700 | } |
| 701 | } |
| 702 | |
| 703 | /** |
| 704 | * Read global counter |
| 705 | * |
| 706 | * Read the table with number of calls for all functions |
| 707 | * |
| 708 | * @param stats vector used to read data from control table. |
| 709 | * @param maps_per_core do I need to read all cores? |
| 710 | */ |
| 711 | static void ebpf_fd_read_global_tables(netdata_idx_t *stats, int maps_per_core) |
| 712 | { |
| 713 | ebpf_read_global_table_stats( |
| 714 | fd_hash_values, |
| 715 | fd_values, |
| 716 | fd_maps[NETDATA_FD_GLOBAL_STATS].map_fd, |
| 717 | maps_per_core, |
| 718 | NETDATA_KEY_CALLS_DO_SYS_OPEN, |
| 719 | NETDATA_FD_COUNTER); |
| 720 | |
| 721 | ebpf_read_global_table_stats( |
| 722 | stats, |
| 723 | fd_values, |
| 724 | fd_maps[NETDATA_FD_CONTROLLER].map_fd, |
| 725 | maps_per_core, |
| 726 | NETDATA_CONTROLLER_PID_TABLE_ADD, |
| 727 | NETDATA_CONTROLLER_END); |
| 728 | } |
| 729 | |
| 730 | /** |
| 731 | * Apps Accumulator |
| 732 | * |
| 733 | * Sum all values read from kernel and store in the first address. |
| 734 | * |
| 735 | * @param out the vector with read values. |
| 736 | * @param maps_per_core do I need to read all cores? |
| 737 | */ |
| 738 | static void fd_apps_accumulator(netdata_fd_stat_t *out, int maps_per_core) |
| 739 | { |
| 740 | int i, end = (maps_per_core) ? ebpf_nprocs : 1; |
| 741 | netdata_fd_stat_t *total = &out[0]; |
| 742 | uint64_t ct = total->ct; |
| 743 | for (i = 1; i < end; i++) { |
| 744 | if (ebpf_plugin_stop()) |
| 745 | break; |
| 746 | |
| 747 | netdata_fd_stat_t *w = &out[i]; |
| 748 | total->open_call += w->open_call; |
| 749 | total->close_call += w->close_call; |
| 750 | total->open_err += w->open_err; |
| 751 | total->close_err += w->close_err; |
| 752 | |
| 753 | if (w->ct > ct) |
| 754 | ct = w->ct; |
| 755 | |
| 756 | if (!total->name[0] && w->name[0]) |
| 757 | strncpyz(total->name, w->name, sizeof(total->name) - 1); |
| 758 | } |
| 759 | } |
| 760 | |
| 761 | /** |
| 762 | * Read APPS table |
| 763 | * |
| 764 | * Read the apps table and store data inside the structure. |
| 765 | * |
| 766 | * @param maps_per_core do I need to read all cores? |
| 767 | */ |
| 768 | static void ebpf_read_fd_apps_table(int maps_per_core) |
| 769 | { |
| 770 | netdata_fd_stat_t *fv = fd_vector; |
| 771 | int fd = fd_maps[NETDATA_FD_PID_STATS].map_fd; |
| 772 | size_t length = sizeof(netdata_fd_stat_t); |
| 773 | if (maps_per_core) |
| 774 | length *= ebpf_nprocs; |
| 775 | |
| 776 | uint32_t key = 0, next_key = 0; |
| 777 | while (bpf_map_get_next_key(fd, &key, &next_key) == 0) { |
| 778 | if (ebpf_plugin_stop()) |
| 779 | break; |
| 780 | |
| 781 | key = next_key; |
| 782 | |
| 783 | if (bpf_map_lookup_elem(fd, &key, fv)) { |
| 784 | netdata_log_error("Failed to lookup PID %u in FD map", key); |
| 785 | goto end_fd_loop; |
| 786 | } |
| 787 | |
| 788 | fd_apps_accumulator(fv, maps_per_core); |
| 789 | |
| 790 | netdata_ebpf_pid_stats_t *local_pid = netdata_ebpf_get_shm_pointer_unsafe(key, NETDATA_EBPF_PIDS_FD_IDX); |
| 791 | if (!local_pid) |
| 792 | goto end_fd_loop; |
| 793 | netdata_publish_fd_stat_t *publish_fd = &local_pid->fd; |
| 794 | |
| 795 | if (kill((pid_t)key, 0) == -1 && errno == ESRCH) { |
| 796 | if (netdata_ebpf_reset_shm_pointer_unsafe(fd, key, NETDATA_EBPF_PIDS_FD_IDX)) |
| 797 | memset(publish_fd, 0, sizeof(*publish_fd)); |
| 798 | } else if (!publish_fd->ct || publish_fd->ct != fv->ct) { |
| 799 | publish_fd->ct = fv->ct; |
| 800 | publish_fd->open_call = fv->open_call; |
| 801 | publish_fd->close_call = fv->close_call; |
| 802 | publish_fd->open_err = fv->open_err; |
| 803 | publish_fd->close_err = fv->close_err; |
| 804 | } |
| 805 | |
| 806 | end_fd_loop: |
| 807 | // We are cleaning to avoid passing data read from one process to other. |
| 808 | memset(fv, 0, length); |
| 809 | } |
| 810 | } |
| 811 | |
| 812 | /** |
| 813 | * Sum PIDs |
| 814 | * |
| 815 | * Sum values for all targets. |
| 816 | * |
| 817 | * @param fd the output |
| 818 | * @param root list of pids |
| 819 | */ |
| 820 | static void ebpf_fd_sum_pids(netdata_fd_stat_t *fd, struct ebpf_pid_on_target *root) |
| 821 | { |
| 822 | memset(fd, 0, sizeof(netdata_fd_stat_t)); |
| 823 | |
| 824 | for (; root; root = root->next) { |
| 825 | uint32_t pid = root->pid; |
| 826 | netdata_ebpf_pid_stats_t *pid_stat = netdata_ebpf_lookup_shm_pointer_unsafe(pid); |
| 827 | if (!pid_stat || !(pid_stat->threads & (1U << (NETDATA_EBPF_PIDS_FD_IDX << 1)))) |
| 828 | continue; |
| 829 | netdata_publish_fd_stat_t *w = &pid_stat->fd; |
| 830 | |
| 831 | fd->open_call += w->open_call; |
| 832 | fd->close_call += w->close_call; |
| 833 | fd->open_err += w->open_err; |
| 834 | fd->close_err += w->close_err; |
| 835 | } |
| 836 | } |
| 837 | |
| 838 | /** |
| 839 | * Resume apps data |
| 840 | */ |
| 841 | void ebpf_fd_resume_apps_data(void) |
| 842 | { |
| 843 | struct ebpf_target *w; |
| 844 | |
| 845 | netdata_mutex_lock(&collect_data_mutex); |
| 846 | for (w = apps_groups_root_target; w; w = w->next) { |
| 847 | if (ebpf_plugin_stop()) |
| 848 | break; |
| 849 | |
| 850 | if (unlikely(!(w->charts_created & (1 << EBPF_MODULE_FD_IDX)))) |
| 851 | continue; |
| 852 | |
| 853 | ebpf_fd_sum_pids(&w->fd, w->root_pid); |
| 854 | } |
| 855 | netdata_mutex_unlock(&collect_data_mutex); |
| 856 | } |
| 857 | |
| 858 | /** |
| 859 | * Update cgroup |
| 860 | * |
| 861 | * Update cgroup data collected per PID. |
| 862 | * |
| 863 | * @param maps_per_core do I need to read all cores? |
| 864 | */ |
| 865 | static void ebpf_update_fd_cgroup(void) |
| 866 | { |
| 867 | ebpf_cgroup_target_t *ect; |
| 868 | |
| 869 | netdata_mutex_lock(&mutex_cgroup_shm); |
| 870 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 871 | if (ebpf_plugin_stop()) |
| 872 | break; |
| 873 | |
| 874 | struct pid_on_target2 *pids; |
| 875 | for (pids = ect->pids; pids; pids = pids->next) { |
| 876 | uint32_t pid = pids->pid; |
| 877 | netdata_publish_fd_stat_t *out = &pids->fd; |
| 878 | |
| 879 | netdata_ebpf_pid_stats_t *pid_stat = netdata_ebpf_lookup_shm_pointer_unsafe(pid); |
| 880 | if (!pid_stat || !(pid_stat->threads & (1U << (NETDATA_EBPF_PIDS_FD_IDX << 1)))) |
| 881 | continue; |
| 882 | |
| 883 | netdata_publish_fd_stat_t *in = &pid_stat->fd; |
| 884 | |
| 885 | memcpy(out, in, sizeof(netdata_publish_fd_stat_t)); |
| 886 | } |
| 887 | } |
| 888 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 889 | } |
| 890 | |
| 891 | /** |
| 892 | * FD thread |
| 893 | * |
| 894 | * Thread used to generate fd charts. |
| 895 | * |
| 896 | * @param ptr a pointer to `struct ebpf_module` |
| 897 | * |
| 898 | * @return It always return NULL |
| 899 | */ |
| 900 | void ebpf_read_fd_thread(void *ptr) |
| 901 | { |
| 902 | ebpf_module_t *em = (ebpf_module_t *)ptr; |
| 903 | |
| 904 | int maps_per_core = em->maps_per_core; |
| 905 | int update_every = em->update_every; |
| 906 | int collect_pid = (em->apps_charts || em->cgroup_charts); |
| 907 | if (!collect_pid) |
| 908 | return; |
| 909 | |
| 910 | int counter = update_every - 1; |
| 911 | |
| 912 | uint32_t lifetime = em->lifetime; |
| 913 | int cgroups = em->cgroup_charts; |
| 914 | uint32_t running_time = 0; |
| 915 | ebpf_set_pid_map_fd(NETDATA_EBPF_PIDS_FD_IDX, fd_maps[NETDATA_FD_PID_STATS].map_fd); |
| 916 | |
| 917 | heartbeat_t hb; |
| 918 | heartbeat_init(&hb, USEC_PER_SEC); |
| 919 | while (!ebpf_plugin_stop() && running_time < lifetime) { |
| 920 | if (ebpf_plugin_stop()) |
| 921 | break; |
| 922 | |
| 923 | heartbeat_next(&hb); |
| 924 | if (ebpf_plugin_stop()) |
| 925 | break; |
| 926 | |
| 927 | if (++counter != update_every) |
| 928 | continue; |
| 929 | |
| 930 | if (!ebpf_shm_sem_wait_or_stop(shm_mutex_ebpf_integration)) { |
| 931 | if (errno != ECANCELED) |
| 932 | netdata_log_error("FD: Failed to wait on semaphore."); |
| 933 | break; |
| 934 | } |
| 935 | ebpf_read_fd_apps_table(maps_per_core); |
| 936 | ebpf_fd_resume_apps_data(); |
| 937 | if (ebpf_plugin_stop()) { |
| 938 | if (sem_post(shm_mutex_ebpf_integration)) |
| 939 | netdata_log_error("FD: Failed to post semaphore."); |
| 940 | break; |
| 941 | } |
| 942 | |
| 943 | if (cgroups && ebpf_cgroup_integration_active_get()) |
| 944 | ebpf_update_fd_cgroup(); |
| 945 | |
| 946 | if (sem_post(shm_mutex_ebpf_integration)) { |
| 947 | netdata_log_error("FD: Failed to post semaphore."); |
| 948 | break; |
| 949 | } |
| 950 | |
| 951 | counter = 0; |
| 952 | |
| 953 | if (ebpf_plugin_stop()) |
| 954 | break; |
| 955 | |
| 956 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 957 | if (running_time && !em->running_time) |
| 958 | running_time = update_every; |
| 959 | else |
| 960 | running_time += update_every; |
| 961 | |
| 962 | em->running_time = running_time; |
| 963 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 964 | } |
| 965 | } |
| 966 | |
| 967 | /** |
| 968 | * Send data to Netdata calling auxiliary functions. |
| 969 | * |
| 970 | * @param em the structure with thread information |
| 971 | * @param root the target list. |
| 972 | */ |
| 973 | void ebpf_fd_send_apps_data(ebpf_module_t *em, struct ebpf_target *root) |
| 974 | { |
| 975 | struct ebpf_target *w; |
| 976 | netdata_mutex_lock(&collect_data_mutex); |
| 977 | for (w = root; w; w = w->next) { |
| 978 | if (ebpf_plugin_stop()) |
| 979 | break; |
| 980 | |
| 981 | if (unlikely(!(w->charts_created & (1 << EBPF_MODULE_FD_IDX)))) |
| 982 | continue; |
| 983 | |
| 984 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, w->clean_name, "_ebpf_file_open"); |
| 985 | write_chart_dimension("calls", w->fd.open_call); |
| 986 | ebpf_write_end_chart(); |
| 987 | |
| 988 | if (em->mode < MODE_ENTRY) { |
| 989 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, w->clean_name, "_ebpf_file_open_error"); |
| 990 | write_chart_dimension("calls", w->fd.open_err); |
| 991 | ebpf_write_end_chart(); |
| 992 | } |
| 993 | |
| 994 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, w->clean_name, "_ebpf_file_closed"); |
| 995 | write_chart_dimension("calls", w->fd.close_call); |
| 996 | ebpf_write_end_chart(); |
| 997 | |
| 998 | if (em->mode < MODE_ENTRY) { |
| 999 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, w->clean_name, "_ebpf_file_close_error"); |
| 1000 | write_chart_dimension("calls", w->fd.close_err); |
| 1001 | ebpf_write_end_chart(); |
| 1002 | } |
| 1003 | } |
| 1004 | netdata_mutex_unlock(&collect_data_mutex); |
| 1005 | } |
| 1006 | |
| 1007 | /** |
| 1008 | * Sum PIDs |
| 1009 | * |
| 1010 | * Sum values for all targets. |
| 1011 | * |
| 1012 | * @param fd structure used to store data |
| 1013 | * @param pids input data |
| 1014 | */ |
| 1015 | static void ebpf_fd_sum_cgroup_pids(netdata_publish_fd_stat_t *fd, struct pid_on_target2 *pids) |
| 1016 | { |
| 1017 | netdata_fd_stat_t accumulator; |
| 1018 | memset(&accumulator, 0, sizeof(accumulator)); |
| 1019 | |
| 1020 | while (pids) { |
| 1021 | netdata_publish_fd_stat_t *w = &pids->fd; |
| 1022 | |
| 1023 | accumulator.open_err += w->open_err; |
| 1024 | accumulator.open_call += w->open_call; |
| 1025 | accumulator.close_call += w->close_call; |
| 1026 | accumulator.close_err += w->close_err; |
| 1027 | |
| 1028 | pids = pids->next; |
| 1029 | } |
| 1030 | |
| 1031 | fd->open_call = accumulator.open_call; |
| 1032 | fd->open_err = accumulator.open_err; |
| 1033 | fd->close_call = accumulator.close_call; |
| 1034 | fd->close_err = accumulator.close_err; |
| 1035 | } |
| 1036 | |
| 1037 | /** |
| 1038 | * Create specific file descriptor charts |
| 1039 | * |
| 1040 | * Create charts for cgroup/application. |
| 1041 | * |
| 1042 | * @param type the chart type. |
| 1043 | * @param em the main thread structure. |
| 1044 | */ |
| 1045 | static void ebpf_create_specific_fd_charts(char *type, ebpf_module_t *em) |
| 1046 | { |
| 1047 | char *label = (!strncmp(type, "cgroup_", 7)) ? &type[7] : type; |
| 1048 | ebpf_create_chart( |
| 1049 | type, |
| 1050 | NETDATA_SYSCALL_APPS_FILE_OPEN, |
| 1051 | "Number of open files", |
| 1052 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1053 | NETDATA_APPS_FILE_GROUP, |
| 1054 | NETDATA_CGROUP_FD_OPEN_CONTEXT, |
| 1055 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1056 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + NETDATA_EBPF_FD_CHART_PRIORITY_CGROUP_OPEN, |
| 1057 | ebpf_create_global_dimension, |
| 1058 | &fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN], |
| 1059 | 1, |
| 1060 | em->update_every, |
| 1061 | NETDATA_EBPF_MODULE_NAME_FD); |
| 1062 | ebpf_create_chart_labels("cgroup_name", label, RRDLABEL_SRC_AUTO); |
| 1063 | ebpf_commit_label(); |
| 1064 | |
| 1065 | if (em->mode < MODE_ENTRY) { |
| 1066 | ebpf_create_chart( |
| 1067 | type, |
| 1068 | NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR, |
| 1069 | "Fails to open files", |
| 1070 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1071 | NETDATA_APPS_FILE_GROUP, |
| 1072 | NETDATA_CGROUP_FD_OPEN_ERR_CONTEXT, |
| 1073 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1074 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + NETDATA_EBPF_FD_CHART_PRIORITY_CGROUP_OPEN_ERR, |
| 1075 | ebpf_create_global_dimension, |
| 1076 | &fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN], |
| 1077 | 1, |
| 1078 | em->update_every, |
| 1079 | NETDATA_EBPF_MODULE_NAME_FD); |
| 1080 | ebpf_create_chart_labels("cgroup_name", label, RRDLABEL_SRC_AUTO); |
| 1081 | ebpf_commit_label(); |
| 1082 | } |
| 1083 | |
| 1084 | ebpf_create_chart( |
| 1085 | type, |
| 1086 | NETDATA_SYSCALL_APPS_FILE_CLOSED, |
| 1087 | "Files closed", |
| 1088 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1089 | NETDATA_APPS_FILE_GROUP, |
| 1090 | NETDATA_CGROUP_FD_CLOSE_CONTEXT, |
| 1091 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1092 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + NETDATA_EBPF_FD_CHART_PRIORITY_CGROUP_CLOSE, |
| 1093 | ebpf_create_global_dimension, |
| 1094 | &fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE], |
| 1095 | 1, |
| 1096 | em->update_every, |
| 1097 | NETDATA_EBPF_MODULE_NAME_FD); |
| 1098 | ebpf_create_chart_labels("cgroup_name", label, RRDLABEL_SRC_AUTO); |
| 1099 | ebpf_commit_label(); |
| 1100 | |
| 1101 | if (em->mode < MODE_ENTRY) { |
| 1102 | ebpf_create_chart( |
| 1103 | type, |
| 1104 | NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR, |
| 1105 | "Fails to close files", |
| 1106 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1107 | NETDATA_APPS_FILE_GROUP, |
| 1108 | NETDATA_CGROUP_FD_CLOSE_ERR_CONTEXT, |
| 1109 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1110 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + NETDATA_EBPF_FD_CHART_PRIORITY_CGROUP_CLOSE_ERR, |
| 1111 | ebpf_create_global_dimension, |
| 1112 | &fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE], |
| 1113 | 1, |
| 1114 | em->update_every, |
| 1115 | NETDATA_EBPF_MODULE_NAME_FD); |
| 1116 | ebpf_create_chart_labels("cgroup_name", label, RRDLABEL_SRC_AUTO); |
| 1117 | ebpf_commit_label(); |
| 1118 | } |
| 1119 | } |
| 1120 | |
| 1121 | /** |
| 1122 | * Obsolete specific file descriptor charts |
| 1123 | * |
| 1124 | * Obsolete charts for cgroup/application. |
| 1125 | * |
| 1126 | * @param type the chart type. |
| 1127 | * @param em the main thread structure. |
| 1128 | */ |
| 1129 | static void ebpf_obsolete_specific_fd_charts(char *type, ebpf_module_t *em) |
| 1130 | { |
| 1131 | ebpf_write_chart_obsolete( |
| 1132 | type, |
| 1133 | NETDATA_SYSCALL_APPS_FILE_OPEN, |
| 1134 | "", |
| 1135 | "Number of open files", |
| 1136 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1137 | NETDATA_APPS_FILE_GROUP, |
| 1138 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1139 | NETDATA_CGROUP_FD_OPEN_CONTEXT, |
| 1140 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + NETDATA_EBPF_FD_CHART_PRIORITY_CGROUP_OPEN, |
| 1141 | em->update_every); |
| 1142 | |
| 1143 | if (em->mode < MODE_ENTRY) { |
| 1144 | ebpf_write_chart_obsolete( |
| 1145 | type, |
| 1146 | NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR, |
| 1147 | "", |
| 1148 | "Fails to open files", |
| 1149 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1150 | NETDATA_APPS_FILE_GROUP, |
| 1151 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1152 | NETDATA_CGROUP_FD_OPEN_ERR_CONTEXT, |
| 1153 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + NETDATA_EBPF_FD_CHART_PRIORITY_CGROUP_OPEN_ERR, |
| 1154 | em->update_every); |
| 1155 | } |
| 1156 | |
| 1157 | ebpf_write_chart_obsolete( |
| 1158 | type, |
| 1159 | NETDATA_SYSCALL_APPS_FILE_CLOSED, |
| 1160 | "", |
| 1161 | "Files closed", |
| 1162 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1163 | NETDATA_APPS_FILE_GROUP, |
| 1164 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1165 | NETDATA_CGROUP_FD_CLOSE_CONTEXT, |
| 1166 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + NETDATA_EBPF_FD_CHART_PRIORITY_CGROUP_CLOSE, |
| 1167 | em->update_every); |
| 1168 | |
| 1169 | if (em->mode < MODE_ENTRY) { |
| 1170 | ebpf_write_chart_obsolete( |
| 1171 | type, |
| 1172 | NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR, |
| 1173 | "", |
| 1174 | "Fails to close files", |
| 1175 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1176 | NETDATA_APPS_FILE_GROUP, |
| 1177 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1178 | NETDATA_CGROUP_FD_CLOSE_ERR_CONTEXT, |
| 1179 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + NETDATA_EBPF_FD_CHART_PRIORITY_CGROUP_CLOSE_ERR, |
| 1180 | em->update_every); |
| 1181 | } |
| 1182 | } |
| 1183 | |
| 1184 | /* |
| 1185 | * Send specific file descriptor data |
| 1186 | * |
| 1187 | * Send data for specific cgroup/apps. |
| 1188 | * |
| 1189 | * @param type chart type |
| 1190 | * @param values structure with values that will be sent to netdata |
| 1191 | */ |
| 1192 | static void ebpf_send_specific_fd_data(char *type, netdata_publish_fd_stat_t *values, ebpf_module_t *em) |
| 1193 | { |
| 1194 | ebpf_write_begin_chart(type, NETDATA_SYSCALL_APPS_FILE_OPEN, ""); |
| 1195 | write_chart_dimension(fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN].name, (long long)values->open_call); |
| 1196 | ebpf_write_end_chart(); |
| 1197 | |
| 1198 | if (em->mode < MODE_ENTRY) { |
| 1199 | ebpf_write_begin_chart(type, NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR, ""); |
| 1200 | write_chart_dimension(fd_publish_aggregated[NETDATA_FD_SYSCALL_OPEN].name, (long long)values->open_err); |
| 1201 | ebpf_write_end_chart(); |
| 1202 | } |
| 1203 | |
| 1204 | ebpf_write_begin_chart(type, NETDATA_SYSCALL_APPS_FILE_CLOSED, ""); |
| 1205 | write_chart_dimension(fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE].name, (long long)values->close_call); |
| 1206 | ebpf_write_end_chart(); |
| 1207 | |
| 1208 | if (em->mode < MODE_ENTRY) { |
| 1209 | ebpf_write_begin_chart(type, NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR, ""); |
| 1210 | write_chart_dimension(fd_publish_aggregated[NETDATA_FD_SYSCALL_CLOSE].name, (long long)values->close_err); |
| 1211 | ebpf_write_end_chart(); |
| 1212 | } |
| 1213 | } |
| 1214 | |
| 1215 | /** |
| 1216 | * Create systemd file descriptor charts |
| 1217 | * |
| 1218 | * Create charts when systemd is enabled |
| 1219 | * |
| 1220 | * @param em the main collector structure |
| 1221 | **/ |
| 1222 | static void ebpf_create_systemd_fd_charts(ebpf_module_t *em) |
| 1223 | { |
| 1224 | static ebpf_systemd_args_t data_open = { |
| 1225 | .title = "Number of open files", |
| 1226 | .units = EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1227 | .family = NETDATA_APPS_FILE_GROUP, |
| 1228 | .charttype = NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1229 | .order = NETDATA_EBPF_FD_CHART_PRIORITY_OPEN, |
| 1230 | .algorithm = EBPF_CHART_ALGORITHM_INCREMENTAL, |
| 1231 | .context = NETDATA_SYSTEMD_FD_OPEN_CONTEXT, |
| 1232 | .module = NETDATA_EBPF_MODULE_NAME_FD, |
| 1233 | .update_every = 0, |
| 1234 | .suffix = NETDATA_SYSCALL_APPS_FILE_OPEN, |
| 1235 | .dimension = "calls"}; |
| 1236 | |
| 1237 | static ebpf_systemd_args_t data_open_error = { |
| 1238 | .title = "Fails to open files", |
| 1239 | .units = EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1240 | .family = NETDATA_APPS_FILE_GROUP, |
| 1241 | .charttype = NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1242 | .order = NETDATA_EBPF_FD_CHART_PRIORITY_OPEN_ERR, |
| 1243 | .algorithm = EBPF_CHART_ALGORITHM_INCREMENTAL, |
| 1244 | .context = NETDATA_SYSTEMD_FD_OPEN_ERR_CONTEXT, |
| 1245 | .module = NETDATA_EBPF_MODULE_NAME_FD, |
| 1246 | .update_every = 0, |
| 1247 | .suffix = NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR, |
| 1248 | .dimension = "calls"}; |
| 1249 | |
| 1250 | static ebpf_systemd_args_t data_close = { |
| 1251 | .title = "Files closed", |
| 1252 | .units = EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1253 | .family = NETDATA_APPS_FILE_GROUP, |
| 1254 | .charttype = NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1255 | .order = NETDATA_EBPF_FD_CHART_PRIORITY_CLOSE, |
| 1256 | .algorithm = EBPF_CHART_ALGORITHM_INCREMENTAL, |
| 1257 | .context = NETDATA_SYSTEMD_FD_CLOSE_CONTEXT, |
| 1258 | .module = NETDATA_EBPF_MODULE_NAME_FD, |
| 1259 | .update_every = 0, |
| 1260 | .suffix = NETDATA_SYSCALL_APPS_FILE_CLOSED, |
| 1261 | .dimension = "calls"}; |
| 1262 | |
| 1263 | static ebpf_systemd_args_t data_close_error = { |
| 1264 | .title = "Fails to close files", |
| 1265 | .units = EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1266 | .family = NETDATA_APPS_FILE_GROUP, |
| 1267 | .charttype = NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1268 | .order = NETDATA_EBPF_FD_CHART_PRIORITY_CLOSE_ERR, |
| 1269 | .algorithm = EBPF_CHART_ALGORITHM_INCREMENTAL, |
| 1270 | .context = NETDATA_SYSTEMD_FD_CLOSE_ERR_CONTEXT, |
| 1271 | .module = NETDATA_EBPF_MODULE_NAME_FD, |
| 1272 | .update_every = 0, |
| 1273 | .suffix = NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR, |
| 1274 | .dimension = "calls"}; |
| 1275 | |
| 1276 | if (!data_open.update_every) { |
| 1277 | data_open.update_every = em->update_every; |
| 1278 | data_open_error.update_every = em->update_every; |
| 1279 | data_close.update_every = em->update_every; |
| 1280 | data_close_error.update_every = em->update_every; |
| 1281 | } |
| 1282 | |
| 1283 | ebpf_cgroup_target_t *w; |
| 1284 | netdata_run_mode_t mode = em->mode; |
| 1285 | for (w = ebpf_cgroup_pids; w; w = w->next) { |
| 1286 | if (ebpf_plugin_stop()) |
| 1287 | break; |
| 1288 | |
| 1289 | if (unlikely(!w->systemd || w->flags & NETDATA_EBPF_SERVICES_HAS_FD_CHART)) |
| 1290 | continue; |
| 1291 | |
| 1292 | data_open.id = data_open_error.id = data_close.id = data_close_error.id = w->name; |
| 1293 | ebpf_create_charts_on_systemd(&data_open); |
| 1294 | ebpf_create_charts_on_systemd(&data_close); |
| 1295 | if (mode < MODE_ENTRY) { |
| 1296 | ebpf_create_charts_on_systemd(&data_open_error); |
| 1297 | ebpf_create_charts_on_systemd(&data_close_error); |
| 1298 | } |
| 1299 | |
| 1300 | w->flags |= NETDATA_EBPF_SERVICES_HAS_FD_CHART; |
| 1301 | } |
| 1302 | } |
| 1303 | |
| 1304 | /** |
| 1305 | * Send Systemd charts |
| 1306 | * |
| 1307 | * Send collected data to Netdata. |
| 1308 | * |
| 1309 | * @param em the main collector structure |
| 1310 | */ |
| 1311 | static void ebpf_send_systemd_fd_charts(ebpf_module_t *em) |
| 1312 | { |
| 1313 | ebpf_cgroup_target_t *ect; |
| 1314 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 1315 | if (ebpf_plugin_stop()) |
| 1316 | break; |
| 1317 | |
| 1318 | if (unlikely(!(ect->flags & NETDATA_EBPF_SERVICES_HAS_FD_CHART))) { |
| 1319 | continue; |
| 1320 | } |
| 1321 | |
| 1322 | ebpf_write_begin_chart(ect->name, NETDATA_SYSCALL_APPS_FILE_OPEN, ""); |
| 1323 | write_chart_dimension("calls", ect->publish_systemd_fd.open_call); |
| 1324 | ebpf_write_end_chart(); |
| 1325 | |
| 1326 | if (em->mode < MODE_ENTRY) { |
| 1327 | ebpf_write_begin_chart(ect->name, NETDATA_SYSCALL_APPS_FILE_OPEN_ERROR, ""); |
| 1328 | write_chart_dimension("calls", ect->publish_systemd_fd.open_err); |
| 1329 | ebpf_write_end_chart(); |
| 1330 | } |
| 1331 | |
| 1332 | ebpf_write_begin_chart(ect->name, NETDATA_SYSCALL_APPS_FILE_CLOSED, ""); |
| 1333 | write_chart_dimension("calls", ect->publish_systemd_fd.close_call); |
| 1334 | ebpf_write_end_chart(); |
| 1335 | |
| 1336 | if (em->mode < MODE_ENTRY) { |
| 1337 | ebpf_write_begin_chart(ect->name, NETDATA_SYSCALL_APPS_FILE_CLOSE_ERROR, ""); |
| 1338 | write_chart_dimension("calls", ect->publish_systemd_fd.close_err); |
| 1339 | ebpf_write_end_chart(); |
| 1340 | } |
| 1341 | } |
| 1342 | } |
| 1343 | |
| 1344 | /** |
| 1345 | * Send data to Netdata calling auxiliary functions. |
| 1346 | * |
| 1347 | * @param em the main collector structure |
| 1348 | */ |
| 1349 | static void ebpf_fd_send_cgroup_data(ebpf_module_t *em) |
| 1350 | { |
| 1351 | netdata_mutex_lock(&mutex_cgroup_shm); |
| 1352 | ebpf_cgroup_target_t *ect; |
| 1353 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 1354 | ebpf_fd_sum_cgroup_pids(&ect->publish_systemd_fd, ect->pids); |
| 1355 | } |
| 1356 | |
| 1357 | if (ebpf_plugin_stop()) { |
| 1358 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 1359 | return; |
| 1360 | } |
| 1361 | |
| 1362 | if (ebpf_cgroup_systemd_enabled_get()) { |
| 1363 | if (ebpf_send_cgroup_chart_get()) { |
| 1364 | ebpf_create_systemd_fd_charts(em); |
| 1365 | } |
| 1366 | |
| 1367 | ebpf_send_systemd_fd_charts(em); |
| 1368 | } |
| 1369 | |
| 1370 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 1371 | if (ebpf_plugin_stop()) |
| 1372 | break; |
| 1373 | |
| 1374 | if (ect->systemd) |
| 1375 | continue; |
| 1376 | |
| 1377 | if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_FD_CHART) && ect->updated) { |
| 1378 | ebpf_create_specific_fd_charts(ect->name, em); |
| 1379 | ect->flags |= NETDATA_EBPF_CGROUP_HAS_FD_CHART; |
| 1380 | } |
| 1381 | |
| 1382 | if (ect->flags & NETDATA_EBPF_CGROUP_HAS_FD_CHART) { |
| 1383 | if (ect->updated) { |
| 1384 | ebpf_send_specific_fd_data(ect->name, &ect->publish_systemd_fd, em); |
| 1385 | } else { |
| 1386 | ebpf_obsolete_specific_fd_charts(ect->name, em); |
| 1387 | ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_FD_CHART; |
| 1388 | } |
| 1389 | } |
| 1390 | } |
| 1391 | |
| 1392 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 1393 | } |
| 1394 | |
| 1395 | /** |
| 1396 | * Main loop for this collector. |
| 1397 | */ |
| 1398 | static void fd_collector(ebpf_module_t *em) |
| 1399 | { |
| 1400 | int cgroups = em->cgroup_charts; |
| 1401 | int update_every = em->update_every; |
| 1402 | int counter = update_every - 1; |
| 1403 | int maps_per_core = em->maps_per_core; |
| 1404 | uint32_t running_time = 0; |
| 1405 | uint32_t lifetime = em->lifetime; |
| 1406 | netdata_idx_t *stats = em->hash_table_stats; |
| 1407 | memset(stats, 0, sizeof(em->hash_table_stats)); |
| 1408 | heartbeat_t hb; |
| 1409 | heartbeat_init(&hb, USEC_PER_SEC); |
| 1410 | while (!ebpf_plugin_stop() && running_time < lifetime) { |
| 1411 | if (ebpf_plugin_stop()) |
| 1412 | break; |
| 1413 | |
| 1414 | heartbeat_next(&hb); |
| 1415 | |
| 1416 | if (ebpf_plugin_stop()) |
| 1417 | break; |
| 1418 | |
| 1419 | if (++counter != update_every) |
| 1420 | continue; |
| 1421 | |
| 1422 | counter = 0; |
| 1423 | netdata_apps_integration_flags_t apps = em->apps_charts; |
| 1424 | ebpf_fd_read_global_tables(stats, maps_per_core); |
| 1425 | |
| 1426 | if (ebpf_plugin_stop()) |
| 1427 | break; |
| 1428 | |
| 1429 | netdata_mutex_lock(&lock); |
| 1430 | |
| 1431 | ebpf_fd_send_data(em); |
| 1432 | |
| 1433 | if (apps & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) |
| 1434 | ebpf_fd_send_apps_data(em, apps_groups_root_target); |
| 1435 | |
| 1436 | if (ebpf_plugin_stop()) { |
| 1437 | netdata_mutex_unlock(&lock); |
| 1438 | break; |
| 1439 | } |
| 1440 | |
| 1441 | if (cgroups && ebpf_cgroup_integration_active_get()) |
| 1442 | ebpf_fd_send_cgroup_data(em); |
| 1443 | |
| 1444 | netdata_mutex_unlock(&lock); |
| 1445 | |
| 1446 | if (ebpf_plugin_stop()) |
| 1447 | break; |
| 1448 | |
| 1449 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 1450 | if (running_time && !em->running_time) |
| 1451 | running_time = update_every; |
| 1452 | else |
| 1453 | running_time += update_every; |
| 1454 | |
| 1455 | em->running_time = running_time; |
| 1456 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 1457 | } |
| 1458 | } |
| 1459 | |
| 1460 | /***************************************************************** |
| 1461 | * |
| 1462 | * CREATE CHARTS |
| 1463 | * |
| 1464 | *****************************************************************/ |
| 1465 | |
| 1466 | /** |
| 1467 | * Create apps charts |
| 1468 | * |
| 1469 | * Call ebpf_create_chart to create the charts on apps submenu. |
| 1470 | * |
| 1471 | * @param em a pointer to the structure with the default values. |
| 1472 | */ |
| 1473 | void ebpf_fd_create_apps_charts(struct ebpf_module *em, void *ptr) |
| 1474 | { |
| 1475 | struct ebpf_target *root = ptr; |
| 1476 | struct ebpf_target *w; |
| 1477 | int update_every = em->update_every; |
| 1478 | for (w = root; w; w = w->next) { |
| 1479 | if (ebpf_plugin_stop()) |
| 1480 | break; |
| 1481 | |
| 1482 | if (unlikely(!w->exposed)) |
| 1483 | continue; |
| 1484 | |
| 1485 | ebpf_write_chart_cmd( |
| 1486 | NETDATA_APP_FAMILY, |
| 1487 | w->clean_name, |
| 1488 | "_ebpf_file_open", |
| 1489 | "Number of open files", |
| 1490 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1491 | NETDATA_APPS_FILE_FDS, |
| 1492 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1493 | "app.ebpf_file_open", |
| 1494 | 20220, |
| 1495 | update_every, |
| 1496 | NETDATA_EBPF_MODULE_NAME_FD); |
| 1497 | ebpf_create_chart_labels("app_group", w->name, RRDLABEL_SRC_AUTO); |
| 1498 | ebpf_commit_label(); |
| 1499 | fprintf(stdout, "DIMENSION calls '' %s 1 1\n", ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); |
| 1500 | |
| 1501 | if (em->mode < MODE_ENTRY) { |
| 1502 | ebpf_write_chart_cmd( |
| 1503 | NETDATA_APP_FAMILY, |
| 1504 | w->clean_name, |
| 1505 | "_ebpf_file_open_error", |
| 1506 | "Fails to open files.", |
| 1507 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1508 | NETDATA_APPS_FILE_FDS, |
| 1509 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1510 | "app.ebpf_file_open_error", |
| 1511 | 20221, |
| 1512 | update_every, |
| 1513 | NETDATA_EBPF_MODULE_NAME_FD); |
| 1514 | ebpf_create_chart_labels("app_group", w->name, RRDLABEL_SRC_AUTO); |
| 1515 | ebpf_commit_label(); |
| 1516 | fprintf(stdout, "DIMENSION calls '' %s 1 1\n", ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); |
| 1517 | } |
| 1518 | |
| 1519 | ebpf_write_chart_cmd( |
| 1520 | NETDATA_APP_FAMILY, |
| 1521 | w->clean_name, |
| 1522 | "_ebpf_file_closed", |
| 1523 | "Files closed.", |
| 1524 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1525 | NETDATA_APPS_FILE_FDS, |
| 1526 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1527 | "app.ebpf_file_closed", |
| 1528 | 20222, |
| 1529 | update_every, |
| 1530 | NETDATA_EBPF_MODULE_NAME_FD); |
| 1531 | ebpf_create_chart_labels("app_group", w->name, RRDLABEL_SRC_AUTO); |
| 1532 | ebpf_commit_label(); |
| 1533 | fprintf(stdout, "DIMENSION calls '' %s 1 1\n", ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); |
| 1534 | |
| 1535 | if (em->mode < MODE_ENTRY) { |
| 1536 | ebpf_write_chart_cmd( |
| 1537 | NETDATA_APP_FAMILY, |
| 1538 | w->clean_name, |
| 1539 | "_ebpf_file_close_error", |
| 1540 | "Fails to close files.", |
| 1541 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1542 | NETDATA_APPS_FILE_FDS, |
| 1543 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 1544 | "app.ebpf_file_close_error", |
| 1545 | NETDATA_EBPF_FD_CHART_PRIORITY_APPS_CLOSE_ERR, |
| 1546 | update_every, |
| 1547 | NETDATA_EBPF_MODULE_NAME_FD); |
| 1548 | ebpf_create_chart_labels("app_group", w->name, RRDLABEL_SRC_AUTO); |
| 1549 | ebpf_commit_label(); |
| 1550 | fprintf(stdout, "DIMENSION calls '' %s 1 1\n", ebpf_algorithms[NETDATA_EBPF_INCREMENTAL_IDX]); |
| 1551 | } |
| 1552 | |
| 1553 | w->charts_created |= 1 << EBPF_MODULE_FD_IDX; |
| 1554 | } |
| 1555 | |
| 1556 | em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; |
| 1557 | } |
| 1558 | |
| 1559 | /** |
| 1560 | * Create global charts |
| 1561 | * |
| 1562 | * Call ebpf_create_chart to create the charts for the collector. |
| 1563 | * |
| 1564 | * @param em a pointer to the structure with the default values. |
| 1565 | */ |
| 1566 | static void ebpf_create_fd_global_charts(ebpf_module_t *em) |
| 1567 | { |
| 1568 | ebpf_create_chart( |
| 1569 | NETDATA_FILESYSTEM_FAMILY, |
| 1570 | NETDATA_FILE_OPEN_CLOSE_COUNT, |
| 1571 | "Open and close calls", |
| 1572 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1573 | NETDATA_FILE_GROUP, |
| 1574 | NETDATA_FS_FILEDESCRIPTOR_CONTEXT, |
| 1575 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1576 | NETDATA_CHART_PRIO_EBPF_FD_CHARTS, |
| 1577 | ebpf_create_global_dimension, |
| 1578 | fd_publish_aggregated, |
| 1579 | NETDATA_FD_SYSCALL_END, |
| 1580 | em->update_every, |
| 1581 | NETDATA_EBPF_MODULE_NAME_FD); |
| 1582 | |
| 1583 | if (em->mode < MODE_ENTRY) { |
| 1584 | ebpf_create_chart( |
| 1585 | NETDATA_FILESYSTEM_FAMILY, |
| 1586 | NETDATA_FILE_OPEN_ERR_COUNT, |
| 1587 | "Open fails", |
| 1588 | EBPF_COMMON_UNITS_CALLS_PER_SEC, |
| 1589 | NETDATA_FILE_GROUP, |
| 1590 | NETDATA_FS_FILEDESCRIPTOR_ERROR_CONTEXT, |
| 1591 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 1592 | NETDATA_CHART_PRIO_EBPF_FD_CHARTS + 1, |
| 1593 | ebpf_create_global_dimension, |
| 1594 | fd_publish_aggregated, |
| 1595 | NETDATA_FD_SYSCALL_END, |
| 1596 | em->update_every, |
| 1597 | NETDATA_EBPF_MODULE_NAME_FD); |
| 1598 | } |
| 1599 | |
| 1600 | fflush(stdout); |
| 1601 | } |
| 1602 | |
| 1603 | /***************************************************************** |
| 1604 | * |
| 1605 | * MAIN THREAD |
| 1606 | * |
| 1607 | *****************************************************************/ |
| 1608 | |
| 1609 | /** |
| 1610 | * Allocate vectors used with this thread. |
| 1611 | * |
| 1612 | * We are not testing the return, because callocz does this and shutdown the software |
| 1613 | * case it was not possible to allocate. |
| 1614 | */ |
| 1615 | static inline void ebpf_fd_allocate_global_vectors(void) |
| 1616 | { |
| 1617 | fd_vector = callocz((size_t)ebpf_nprocs, sizeof(netdata_fd_stat_t)); |
| 1618 | fd_values = callocz((size_t)ebpf_nprocs, sizeof(netdata_idx_t)); |
| 1619 | } |
| 1620 | |
| 1621 | /* |
| 1622 | * Load BPF |
| 1623 | * |
| 1624 | * Load BPF files. |
| 1625 | * |
| 1626 | * @param em the structure with configuration |
| 1627 | */ |
| 1628 | static int ebpf_fd_load_bpf(ebpf_module_t *em) |
| 1629 | { |
| 1630 | #ifdef LIBBPF_MAJOR_VERSION |
| 1631 | ebpf_define_map_type(fd_maps, em->maps_per_core, running_on_kernel); |
| 1632 | #endif |
| 1633 | |
| 1634 | int ret = 0; |
| 1635 | ebpf_adjust_apps_cgroup(em, em->targets[NETDATA_FD_SYSCALL_OPEN].mode); |
| 1636 | if (em->load & EBPF_LOAD_LEGACY) { |
| 1637 | em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); |
| 1638 | if (!em->probe_links) { |
| 1639 | ret = -1; |
| 1640 | } |
| 1641 | } |
| 1642 | #ifdef LIBBPF_MAJOR_VERSION |
| 1643 | else { |
| 1644 | fd_bpf_obj = fd_bpf__open(); |
| 1645 | if (!fd_bpf_obj) |
| 1646 | ret = -1; |
| 1647 | else { |
| 1648 | ret = ebpf_fd_load_and_attach(fd_bpf_obj, em); |
| 1649 | if (ret) { |
| 1650 | fd_bpf__destroy(fd_bpf_obj); |
| 1651 | fd_bpf_obj = NULL; |
| 1652 | } |
| 1653 | } |
| 1654 | } |
| 1655 | #endif |
| 1656 | |
| 1657 | if (ret) |
| 1658 | netdata_log_error("%s %s", EBPF_DEFAULT_ERROR_MSG, em->info.thread_name); |
| 1659 | |
| 1660 | return ret; |
| 1661 | } |
| 1662 | |
| 1663 | /** |
| 1664 | * Directory Cache thread |
| 1665 | * |
| 1666 | * Thread used to make dcstat thread |
| 1667 | * |
| 1668 | * @param ptr a pointer to `struct ebpf_module` |
| 1669 | * |
| 1670 | * @return It always returns NULL |
| 1671 | */ |
| 1672 | void ebpf_fd_thread(void *ptr) |
| 1673 | { |
| 1674 | ebpf_module_t *em = (ebpf_module_t *)ptr; |
| 1675 | |
| 1676 | fd_safe_clean = false; |
| 1677 | |
| 1678 | CLEANUP_FUNCTION_REGISTER(ebpf_fd_exit) cleanup_ptr = em; |
| 1679 | |
| 1680 | if (!ebpf_module_thread_has_valid_state(em)) { |
| 1681 | goto endfd; |
| 1682 | } |
| 1683 | |
| 1684 | em->maps = fd_maps; |
| 1685 | |
| 1686 | #ifdef LIBBPF_MAJOR_VERSION |
| 1687 | ebpf_adjust_thread_load(em, default_btf); |
| 1688 | #endif |
| 1689 | if (ebpf_fd_load_bpf(em)) { |
| 1690 | goto endfd; |
| 1691 | } |
| 1692 | ebpf_mark_program_loaded(); |
| 1693 | |
| 1694 | ebpf_fd_allocate_global_vectors(); |
| 1695 | |
| 1696 | int algorithms[NETDATA_FD_SYSCALL_END] = {NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX}; |
| 1697 | |
| 1698 | ebpf_global_labels( |
| 1699 | fd_aggregated_data, fd_publish_aggregated, fd_dimension_names, fd_id_names, algorithms, NETDATA_FD_SYSCALL_END); |
| 1700 | |
| 1701 | netdata_mutex_lock(&lock); |
| 1702 | ebpf_create_fd_global_charts(em); |
| 1703 | ebpf_update_stats(&plugin_statistics, em); |
| 1704 | ebpf_update_kernel_memory_with_vector(&plugin_statistics, em->maps, EBPF_ACTION_STAT_ADD); |
| 1705 | |
| 1706 | netdata_mutex_unlock(&lock); |
| 1707 | |
| 1708 | ebpf_read_fd.thread = nd_thread_create(ebpf_read_fd.name, NETDATA_THREAD_OPTION_DEFAULT, ebpf_read_fd_thread, em); |
| 1709 | |
| 1710 | fd_safe_clean = true; |
| 1711 | fd_collector(em); |
| 1712 | |
| 1713 | endfd: |
| 1714 | ebpf_update_disabled_plugin_stats(em); |
| 1715 | } |