| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | #include "aclk_rx_msgs.h" |
| 4 | |
| 5 | #include "aclk_query_queue.h" |
| 6 | #include "aclk.h" |
| 7 | #include "aclk_capas.h" |
| 8 | #include "aclk_query.h" |
| 9 | #include "mqtt_websockets/aclk_mqtt_workers.h" |
| 10 | |
| 11 | #include "schema-wrappers/proto_2_json.h" |
| 12 | |
| 13 | #define ACLK_V2_PAYLOAD_SEPARATOR "\x0D\x0A\x0D\x0A" |
| 14 | |
| 15 | #define ACLK_V_COMPRESSION 2 |
| 16 | |
| 17 | struct aclk_request { |
| 18 | bool has_type; |
| 19 | bool is_http; |
| 20 | // Heap-allocated string fields below are owned by the local instance in |
| 21 | // aclk_handle_cloud_cmd_message. On the v2 success path, msg_id and |
| 22 | // callback_topic are transferred into the query and released by |
| 23 | // aclk_query_free; payload is not consumed by v2 (the HTTP body is |
| 24 | // re-derived from the raw frame) and must be freed by the caller on both |
| 25 | // success and error paths. Any new owned field added here must extend |
| 26 | // both cleanup paths to preserve this invariant. |
| 27 | char *msg_id; |
| 28 | char *callback_topic; |
| 29 | char *payload; |
| 30 | int version; |
| 31 | int timeout; |
| 32 | int min_version; |
| 33 | int max_version; |
| 34 | }; |
| 35 | |
| 36 | static int cloud_to_agent_parse(JSON_ENTRY *e) |
| 37 | { |
| 38 | struct aclk_request *data = e->callback_data; |
| 39 | |
| 40 | switch (e->type) { |
| 41 | case JSON_OBJECT: |
| 42 | case JSON_ARRAY: |
| 43 | break; |
| 44 | case JSON_STRING: |
| 45 | if (!strcmp(e->name, "msg-id")) { |
| 46 | data->msg_id = e->data.string ? strdupz(e->data.string) : NULL; |
| 47 | break; |
| 48 | } |
| 49 | if (!strcmp(e->name, "type")) { |
| 50 | data->has_type = true; |
| 51 | data->is_http = (e->data.string && (strcmp(e->data.string, "http") == 0)); |
| 52 | break; |
| 53 | } |
| 54 | if (!strcmp(e->name, "callback-topic")) { |
| 55 | data->callback_topic = e->data.string ? strdupz(e->data.string) : NULL; |
| 56 | break; |
| 57 | } |
| 58 | if (!strcmp(e->name, "payload")) { |
| 59 | if (likely(e->data.string)) { |
| 60 | size_t len = strlen(e->data.string); |
| 61 | data->payload = mallocz(len+1); |
| 62 | if (!url_decode_r(data->payload, e->data.string, len + 1)) |
| 63 | strcpy(data->payload, e->data.string); |
| 64 | } |
| 65 | break; |
| 66 | } |
| 67 | break; |
| 68 | case JSON_NUMBER: |
| 69 | if (!strcmp(e->name, "version")) { |
| 70 | data->version = (int)e->data.number; |
| 71 | break; |
| 72 | } |
| 73 | if (!strcmp(e->name, "timeout")) { |
| 74 | data->timeout = (int)e->data.number; |
| 75 | break; |
| 76 | } |
| 77 | if (!strcmp(e->name, "min-version")) { |
| 78 | data->min_version = (int)e->data.number; |
| 79 | break; |
| 80 | } |
| 81 | if (!strcmp(e->name, "max-version")) { |
| 82 | data->max_version = (int)e->data.number; |
| 83 | break; |
| 84 | } |
| 85 | |
| 86 | break; |
| 87 | |
| 88 | case JSON_BOOLEAN: |
| 89 | break; |
| 90 | |
| 91 | case JSON_NULL: |
| 92 | break; |
| 93 | } |
| 94 | return 0; |
| 95 | } |
| 96 | |
| 97 | static inline int aclk_extract_v2_data(char *payload, char **data) |
| 98 | { |
| 99 | char* ptr = strstr(payload, ACLK_V2_PAYLOAD_SEPARATOR); |
| 100 | if(!ptr) |
| 101 | return 1; |
| 102 | ptr += strlen(ACLK_V2_PAYLOAD_SEPARATOR); |
| 103 | *data = strdupz(ptr); |
| 104 | return 0; |
| 105 | } |
| 106 | |
| 107 | static inline int aclk_v2_payload_get_query(const char *payload, char **query_url) |
| 108 | { |
| 109 | const char *start, *end; |
| 110 | |
| 111 | if(strncmp(payload, "GET /", 5) == 0 || strncmp(payload, "PUT /", 5) == 0) |
| 112 | start = payload + 4; |
| 113 | else if(strncmp(payload, "POST /", 6) == 0) |
| 114 | start = payload + 5; |
| 115 | else if(strncmp(payload, "DELETE /", 8) == 0) |
| 116 | start = payload + 7; |
| 117 | else { |
| 118 | errno_clear(); |
| 119 | netdata_log_error("Only accepting requests that start with GET, POST, PUT, DELETE from CLOUD."); |
| 120 | return 1; |
| 121 | } |
| 122 | |
| 123 | if(!(end = strstr(payload, HTTP_1_1 HTTP_ENDL))) { |
| 124 | errno_clear(); |
| 125 | netdata_log_error("Doesn't look like HTTP GET request."); |
| 126 | return 1; |
| 127 | } |
| 128 | |
| 129 | *query_url = mallocz((end - start) + 1); |
| 130 | strncpyz(*query_url, start, end - start); |
| 131 | |
| 132 | return 0; |
| 133 | } |
| 134 | |
| 135 | static int aclk_handle_cloud_http_request_v2(struct aclk_request *cloud_to_agent, char *raw_payload) |
| 136 | { |
| 137 | errno_clear(); |
| 138 | if (cloud_to_agent->version < ACLK_V_COMPRESSION) { |
| 139 | netdata_log_error( |
| 140 | "This handler cannot reply to request with version older than %d, received %d.", |
| 141 | ACLK_V_COMPRESSION, |
| 142 | cloud_to_agent->version); |
| 143 | return 1; |
| 144 | } |
| 145 | |
| 146 | aclk_query_t *query = aclk_query_new(HTTP_API_V2); |
| 147 | |
| 148 | if (unlikely(aclk_extract_v2_data(raw_payload, &query->data.http_api_v2.payload))) { |
| 149 | netdata_log_error("Error extracting payload expected after the JSON dictionary."); |
| 150 | goto error; |
| 151 | } |
| 152 | |
| 153 | if (unlikely(aclk_v2_payload_get_query(query->data.http_api_v2.payload, &query->dedup_id))) { |
| 154 | netdata_log_error("Could not extract payload from query"); |
| 155 | goto error; |
| 156 | } |
| 157 | |
| 158 | if (unlikely(!cloud_to_agent->callback_topic)) { |
| 159 | netdata_log_error("Missing callback_topic"); |
| 160 | goto error; |
| 161 | } |
| 162 | |
| 163 | if (unlikely(!cloud_to_agent->msg_id)) { |
| 164 | netdata_log_error("Missing msg_id"); |
| 165 | goto error; |
| 166 | } |
| 167 | |
| 168 | // aclk_queue_query takes ownership of data pointer |
| 169 | query->callback_topic = cloud_to_agent->callback_topic; |
| 170 | query->timeout = cloud_to_agent->timeout; |
| 171 | // for clarity and code readability as when we process the request |
| 172 | // it would be strange to get URL from `dedup_id` |
| 173 | query->data.http_api_v2.query = query->dedup_id; |
| 174 | query->msg_id = cloud_to_agent->msg_id; |
| 175 | aclk_execute_query(query); |
| 176 | return 0; |
| 177 | |
| 178 | error: |
| 179 | aclk_query_free(query); |
| 180 | return 1; |
| 181 | } |
| 182 | |
| 183 | int aclk_handle_cloud_cmd_message(char *payload) |
| 184 | { |
| 185 | struct aclk_request cloud_to_agent; |
| 186 | memset(&cloud_to_agent, 0, sizeof(struct aclk_request)); |
| 187 | |
| 188 | if (unlikely(!payload)) { |
| 189 | error_report("ACLK incoming 'cmd' message is empty"); |
| 190 | return 1; |
| 191 | } |
| 192 | |
| 193 | netdata_log_debug(D_ACLK, "ACLK incoming 'cmd' message (%s)", payload); |
| 194 | |
| 195 | int rc = json_parse(payload, &cloud_to_agent, cloud_to_agent_parse); |
| 196 | |
| 197 | if (unlikely(rc != JSON_OK)) { |
| 198 | error_report("Malformed json request (%s)", payload); |
| 199 | goto err_cleanup; |
| 200 | } |
| 201 | |
| 202 | if (!cloud_to_agent.has_type) { |
| 203 | error_report("Cloud message is missing compulsory key \"type\""); |
| 204 | goto err_cleanup; |
| 205 | } |
| 206 | |
| 207 | // Originally we were expecting to have multiple types of 'cmd' message, |
| 208 | // but after the new protocol was designed we will ever only have 'http' |
| 209 | if (!cloud_to_agent.is_http) { |
| 210 | error_report("Only 'http' cmd message is supported"); |
| 211 | goto err_cleanup; |
| 212 | } |
| 213 | |
| 214 | if (likely(!aclk_handle_cloud_http_request_v2(&cloud_to_agent, payload))) { |
| 215 | // aclk_handle_cloud_http_request_v2 takes ownership of msg_id and |
| 216 | // callback_topic on success. The JSON-parsed payload field is not |
| 217 | // consumed by v2 (the HTTP body comes from the raw frame), so free |
| 218 | // it here to avoid leaking when the cmd JSON included a "payload" key. |
| 219 | freez(cloud_to_agent.payload); |
| 220 | return 0; |
| 221 | } |
| 222 | |
| 223 | err_cleanup: |
| 224 | if (cloud_to_agent.payload) |
| 225 | freez(cloud_to_agent.payload); |
| 226 | if (cloud_to_agent.msg_id) |
| 227 | freez(cloud_to_agent.msg_id); |
| 228 | if (cloud_to_agent.callback_topic) |
| 229 | freez(cloud_to_agent.callback_topic); |
| 230 | |
| 231 | return 1; |
| 232 | } |
| 233 | |
| 234 | typedef uint32_t simple_hash_t; |
| 235 | typedef int(*rx_msg_handler)(const char *msg, size_t msg_len); |
| 236 | |
| 237 | int handle_old_proto_cmd(const char *msg, size_t msg_len) |
| 238 | { |
| 239 | // msg is binary payload in all other cases |
| 240 | // however in this message from old legacy cloud |
| 241 | // we have to convert it to C string |
| 242 | char *str = mallocz(msg_len+1); |
| 243 | memcpy(str, msg, msg_len); |
| 244 | str[msg_len] = 0; |
| 245 | int rc = aclk_handle_cloud_cmd_message(str); |
| 246 | freez(str); |
| 247 | return rc; |
| 248 | } |
| 249 | |
| 250 | int create_node_instance_result(const char *msg, size_t msg_len) |
| 251 | { |
| 252 | node_instance_creation_result_t res = parse_create_node_instance_result(msg, msg_len); |
| 253 | if (!res.machine_guid || !res.node_id) { |
| 254 | error_report("Error parsing CreateNodeInstanceResult"); |
| 255 | freez(res.machine_guid); |
| 256 | freez(res.node_id); |
| 257 | return 1; |
| 258 | } |
| 259 | |
| 260 | netdata_log_debug(D_ACLK, "CreateNodeInstanceResult: guid:%s nodeid:%s", res.machine_guid, res.node_id); |
| 261 | |
| 262 | aclk_query_t *query = aclk_query_new(CREATE_NODE_INSTANCE); |
| 263 | |
| 264 | query->data.node_id = res.node_id; // Will be freed on query free |
| 265 | query->machine_guid = res.machine_guid; // Will be freed on query free |
| 266 | aclk_add_job(query); |
| 267 | return 0; |
| 268 | } |
| 269 | |
| 270 | int send_node_instances(const char *msg, size_t msg_len) |
| 271 | { |
| 272 | UNUSED(msg); |
| 273 | UNUSED(msg_len); |
| 274 | aclk_query_t *query = aclk_query_new(SEND_NODE_INSTANCES); |
| 275 | aclk_add_job(query); |
| 276 | return 0; |
| 277 | } |
| 278 | |
| 279 | int stream_charts_and_dimensions(const char *msg, size_t msg_len) |
| 280 | { |
| 281 | UNUSED(msg); |
| 282 | UNUSED(msg_len); |
| 283 | error_report("Received obsolete StreamChartsAndDimensions msg"); |
| 284 | return 0; |
| 285 | } |
| 286 | |
| 287 | int charts_and_dimensions_ack(const char *msg, size_t msg_len) |
| 288 | { |
| 289 | UNUSED(msg); |
| 290 | UNUSED(msg_len); |
| 291 | error_report("Received obsolete StreamChartsAndDimensionsAck msg"); |
| 292 | return 0; |
| 293 | } |
| 294 | |
| 295 | int update_chart_configs(const char *msg, size_t msg_len) |
| 296 | { |
| 297 | UNUSED(msg); |
| 298 | UNUSED(msg_len); |
| 299 | error_report("Received obsolete UpdateChartConfigs msg"); |
| 300 | return 0; |
| 301 | } |
| 302 | |
| 303 | int start_alarm_streaming(const char *msg, size_t msg_len) |
| 304 | { |
| 305 | struct start_alarm_streaming res = parse_start_alarm_streaming(msg, msg_len); |
| 306 | if (!res.node_id) { |
| 307 | netdata_log_error("Error parsing StartAlarmStreaming"); |
| 308 | return 1; |
| 309 | } |
| 310 | aclk_query_t *query = aclk_query_new(ALERT_START_STREAMING); |
| 311 | query->data.node_id = res.node_id; // Will be freed on query free |
| 312 | query->version = res.version; |
| 313 | aclk_add_job(query); |
| 314 | return 0; |
| 315 | } |
| 316 | |
| 317 | int send_alarm_checkpoint(const char *msg, size_t msg_len) |
| 318 | { |
| 319 | struct send_alarm_checkpoint sac = parse_send_alarm_checkpoint(msg, msg_len); |
| 320 | if (!sac.node_id || !sac.claim_id) { |
| 321 | netdata_log_error("Error parsing SendAlarmCheckpoint"); |
| 322 | freez(sac.node_id); |
| 323 | freez(sac.claim_id); |
| 324 | return 1; |
| 325 | } |
| 326 | aclk_query_t *query = aclk_query_new(ALERT_CHECKPOINT); |
| 327 | query->data.node_id = sac.node_id; // Will be freed on query free |
| 328 | query->claim_id = sac.claim_id; |
| 329 | query->version = sac.version; |
| 330 | aclk_add_job(query); |
| 331 | return 0; |
| 332 | } |
| 333 | |
| 334 | int send_alarm_configuration(const char *msg, size_t msg_len) |
| 335 | { |
| 336 | char *config_hash = parse_send_alarm_configuration(msg, msg_len); |
| 337 | if (!config_hash || !*config_hash) { |
| 338 | netdata_log_error("Error parsing SendAlarmConfiguration"); |
| 339 | freez(config_hash); |
| 340 | return 1; |
| 341 | } |
| 342 | aclk_send_alert_configuration(config_hash); |
| 343 | freez(config_hash); |
| 344 | return 0; |
| 345 | } |
| 346 | |
| 347 | int send_alarm_snapshot(const char *msg, size_t msg_len) |
| 348 | { |
| 349 | struct send_alarm_snapshot *sas = parse_send_alarm_snapshot(msg, msg_len); |
| 350 | if (!sas->node_id || !sas->claim_id || !sas->snapshot_uuid) { |
| 351 | netdata_log_error("Error parsing SendAlarmSnapshot"); |
| 352 | destroy_send_alarm_snapshot(sas); |
| 353 | return 1; |
| 354 | } |
| 355 | aclk_query_t *query = aclk_query_new(ALERT_CHECKPOINT); |
| 356 | query->data.node_id = sas->node_id; // Will be freed on query free |
| 357 | query->claim_id = sas->claim_id; // Will be freed on query free |
| 358 | query->version = 0; // force snapshot |
| 359 | aclk_add_job(query); |
| 360 | destroy_send_alarm_snapshot(sas); |
| 361 | return 0; |
| 362 | } |
| 363 | |
| 364 | int handle_disconnect_req(const char *msg, size_t msg_len) |
| 365 | { |
| 366 | struct disconnect_cmd *cmd = parse_disconnect_cmd(msg, msg_len); |
| 367 | if (!cmd) |
| 368 | return 1; |
| 369 | if (cmd->permaban) { |
| 370 | netdata_log_error("Cloud Banned This Agent!"); |
| 371 | aclk_disable_runtime = 1; |
| 372 | } |
| 373 | netdata_log_info("Cloud requested disconnect (EC=%u, \"%s\")", (unsigned int)cmd->error_code, cmd->error_description); |
| 374 | if (cmd->reconnect_after_s > 0) { |
| 375 | aclk_block_until = now_monotonic_sec() + cmd->reconnect_after_s; |
| 376 | netdata_log_info( |
| 377 | "Cloud asks not to reconnect for %u seconds. We shall honor that request", |
| 378 | (unsigned int)cmd->reconnect_after_s); |
| 379 | } |
| 380 | disconnect_req = ACLK_CLOUD_DISCONNECT; |
| 381 | freez(cmd->error_description); |
| 382 | freez(cmd); |
| 383 | return 0; |
| 384 | } |
| 385 | |
| 386 | int contexts_checkpoint(const char *msg, size_t msg_len) |
| 387 | { |
| 388 | aclk_ctx_based = 1; |
| 389 | |
| 390 | struct ctxs_checkpoint *cmd = parse_ctxs_checkpoint(msg, msg_len); |
| 391 | if (!cmd) |
| 392 | return 1; |
| 393 | |
| 394 | aclk_query_t *query = aclk_query_new(CTX_CHECKPOINT); |
| 395 | query->data.payload = cmd; |
| 396 | aclk_add_job(query); |
| 397 | return 0; |
| 398 | } |
| 399 | |
| 400 | int stop_streaming_contexts(const char *msg, size_t msg_len) |
| 401 | { |
| 402 | if (!aclk_ctx_based) { |
| 403 | error_report("Received StopStreamingContexts message but context based communication was not enabled (Cloud violated the protocol). Ignoring message"); |
| 404 | return 1; |
| 405 | } |
| 406 | |
| 407 | struct stop_streaming_ctxs *cmd = parse_stop_streaming_ctxs(msg, msg_len); |
| 408 | if (!cmd) |
| 409 | return 1; |
| 410 | |
| 411 | aclk_query_t *query = aclk_query_new(CTX_STOP_STREAMING); |
| 412 | query->data.payload = cmd; |
| 413 | aclk_add_job(query); |
| 414 | return 0; |
| 415 | } |
| 416 | |
| 417 | int cancel_pending_req(const char *msg, size_t msg_len) |
| 418 | { |
| 419 | struct aclk_cancel_pending_req cmd = {.request_id = NULL, .trace_id = NULL}; |
| 420 | if(parse_cancel_pending_req(msg, msg_len, &cmd)) { |
| 421 | error_report("Error parsing CancelPendingReq"); |
| 422 | return 1; |
| 423 | } |
| 424 | |
| 425 | nd_log(NDLS_ACCESS, NDLP_NOTICE, "ACLK CancelPendingRequest REQ: %s, cloud trace-id: %s", cmd.request_id, cmd.trace_id); |
| 426 | |
| 427 | if (mark_pending_req_cancelled(cmd.request_id)) |
| 428 | error_report("CancelPending Request for %s failed. No such pending request.", cmd.request_id); |
| 429 | |
| 430 | free_cancel_pending_req(&cmd); |
| 431 | return 0; |
| 432 | } |
| 433 | |
| 434 | typedef struct { |
| 435 | const char *name; |
| 436 | simple_hash_t name_hash; |
| 437 | rx_msg_handler fnc; |
| 438 | } new_cloud_rx_msg_t; |
| 439 | |
| 440 | new_cloud_rx_msg_t rx_msgs[] = { |
| 441 | { .name = "cmd", .name_hash = 0, .fnc = handle_old_proto_cmd }, |
| 442 | { .name = "CreateNodeInstanceResult", .name_hash = 0, .fnc = create_node_instance_result }, // async |
| 443 | { .name = "SendNodeInstances", .name_hash = 0, .fnc = send_node_instances }, // async |
| 444 | { .name = "StreamChartsAndDimensions", .name_hash = 0, .fnc = stream_charts_and_dimensions }, // unused |
| 445 | { .name = "ChartsAndDimensionsAck", .name_hash = 0, .fnc = charts_and_dimensions_ack }, // unused |
| 446 | { .name = "UpdateChartConfigs", .name_hash = 0, .fnc = update_chart_configs }, // unused |
| 447 | { .name = "StartAlarmStreaming", .name_hash = 0, .fnc = start_alarm_streaming }, // async |
| 448 | { .name = "SendAlarmCheckpoint", .name_hash = 0, .fnc = send_alarm_checkpoint }, // async |
| 449 | { .name = "SendAlarmConfiguration", .name_hash = 0, .fnc = send_alarm_configuration }, // async |
| 450 | { .name = "SendAlarmSnapshot", .name_hash = 0, .fnc = send_alarm_snapshot }, // shouldn't be used |
| 451 | { .name = "DisconnectReq", .name_hash = 0, .fnc = handle_disconnect_req }, |
| 452 | { .name = "ContextsCheckpoint", .name_hash = 0, .fnc = contexts_checkpoint }, // async |
| 453 | { .name = "StopStreamingContexts", .name_hash = 0, .fnc = stop_streaming_contexts }, // async |
| 454 | { .name = "CancelPendingRequest", .name_hash = 0, .fnc = cancel_pending_req }, |
| 455 | { .name = NULL, .name_hash = 0, .fnc = NULL }, |
| 456 | }; |
| 457 | |
| 458 | new_cloud_rx_msg_t *find_rx_handler_by_hash(simple_hash_t hash) |
| 459 | { |
| 460 | // we can afford to not compare strings after hash match |
| 461 | // because we check for collisions at initialization in |
| 462 | // aclk_init_rx_msg_handlers() |
| 463 | for (int i = 0; rx_msgs[i].fnc; i++) { |
| 464 | if (rx_msgs[i].name_hash == hash) |
| 465 | return &rx_msgs[i]; |
| 466 | } |
| 467 | return NULL; |
| 468 | } |
| 469 | |
| 470 | void aclk_init_rx_msg_handlers(void) |
| 471 | { |
| 472 | int i; |
| 473 | for (i = 0; rx_msgs[i].fnc; i++) { |
| 474 | simple_hash_t hash = simple_hash(rx_msgs[i].name); |
| 475 | new_cloud_rx_msg_t *hdl = find_rx_handler_by_hash(hash); |
| 476 | if (unlikely(hdl)) { |
| 477 | // the list of message names changes only by changing |
| 478 | // the source code, therefore fatal is appropriate |
| 479 | fatal("Hash collision. Choose better hash. Added '%s' clashes with existing '%s'", rx_msgs[i].name, hdl->name); |
| 480 | } |
| 481 | rx_msgs[i].name_hash = hash; |
| 482 | } |
| 483 | } |
| 484 | |
| 485 | void aclk_handle_new_cloud_msg(const char *message_type, const char *msg, size_t msg_len, const char *topic __maybe_unused) |
| 486 | { |
| 487 | new_cloud_rx_msg_t *msg_descriptor = find_rx_handler_by_hash(simple_hash(message_type)); |
| 488 | netdata_log_debug(D_ACLK, "Got message named '%s' from cloud", message_type); |
| 489 | if (unlikely(!msg_descriptor)) { |
| 490 | netdata_log_error("Do not know how to handle message of type '%s'. Ignoring", message_type); |
| 491 | return; |
| 492 | } |
| 493 | |
| 494 | if (aclklog_enabled) { |
| 495 | if (!strncmp(message_type, "cmd", strlen("cmd"))) { |
| 496 | log_aclk_message_bin(msg, msg_len, 0, topic, msg_descriptor->name); |
| 497 | } else { |
| 498 | char *json = protomsg_to_json(msg, msg_len, msg_descriptor->name); |
| 499 | log_aclk_message_bin(json, strlen(json), 0, topic, msg_descriptor->name); |
| 500 | freez(json); |
| 501 | } |
| 502 | } |
| 503 | |
| 504 | if (msg_descriptor->fnc(msg, msg_len)) { |
| 505 | netdata_log_error("Error processing message of type '%s'", message_type); |
| 506 | return; |
| 507 | } |
| 508 | } |