Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 1 | /* |
| 2 | * External health-checks functions. |
| 3 | * |
| 4 | * Copyright 2000-2009,2020 Willy Tarreau <w@1wt.eu> |
| 5 | * Copyright 2014 Horms Solutions Ltd, Simon Horman <horms@verge.net.au> |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or |
| 8 | * modify it under the terms of the GNU General Public License |
| 9 | * as published by the Free Software Foundation; either version |
| 10 | * 2 of the License, or (at your option) any later version. |
| 11 | * |
| 12 | */ |
| 13 | |
| 14 | #include <sys/resource.h> |
| 15 | #include <sys/socket.h> |
| 16 | #include <sys/types.h> |
| 17 | #include <sys/wait.h> |
| 18 | #include <assert.h> |
| 19 | #include <ctype.h> |
| 20 | #include <errno.h> |
| 21 | #include <fcntl.h> |
| 22 | #include <signal.h> |
| 23 | #include <stdarg.h> |
| 24 | #include <stdio.h> |
| 25 | #include <stdlib.h> |
| 26 | #include <string.h> |
| 27 | #include <time.h> |
| 28 | #include <unistd.h> |
| 29 | |
| 30 | #include <haproxy/api.h> |
| 31 | #include <haproxy/cfgparse.h> |
| 32 | #include <haproxy/check.h> |
Willy Tarreau | 36979d9 | 2020-06-05 17:27:29 +0200 | [diff] [blame] | 33 | #include <haproxy/errors.h> |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 34 | #include <haproxy/global.h> |
| 35 | #include <haproxy/list.h> |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 36 | #include <haproxy/proxy.h> |
| 37 | #include <haproxy/server.h> |
| 38 | #include <haproxy/signal.h> |
| 39 | #include <haproxy/task.h> |
| 40 | #include <haproxy/thread.h> |
| 41 | #include <haproxy/time.h> |
| 42 | #include <haproxy/tools.h> |
| 43 | |
| 44 | |
| 45 | static struct list pid_list = LIST_HEAD_INIT(pid_list); |
| 46 | static struct pool_head *pool_head_pid_list; |
| 47 | __decl_spinlock(pid_list_lock); |
| 48 | |
| 49 | struct extcheck_env { |
| 50 | char *name; /* environment variable name */ |
| 51 | int vmaxlen; /* value maximum length, used to determine the required memory allocation */ |
| 52 | }; |
| 53 | |
| 54 | /* environment variables memory requirement for different types of data */ |
| 55 | #define EXTCHK_SIZE_EVAL_INIT 0 /* size determined during the init phase, |
| 56 | * such environment variables are not updatable. */ |
| 57 | #define EXTCHK_SIZE_ULONG 20 /* max string length for an unsigned long value */ |
| 58 | #define EXTCHK_SIZE_UINT 11 /* max string length for an unsigned int value */ |
| 59 | #define EXTCHK_SIZE_ADDR INET6_ADDRSTRLEN+1 /* max string length for an address */ |
| 60 | |
| 61 | /* external checks environment variables */ |
| 62 | enum { |
| 63 | EXTCHK_PATH = 0, |
| 64 | |
| 65 | /* Proxy specific environment variables */ |
| 66 | EXTCHK_HAPROXY_PROXY_NAME, /* the backend name */ |
| 67 | EXTCHK_HAPROXY_PROXY_ID, /* the backend id */ |
| 68 | EXTCHK_HAPROXY_PROXY_ADDR, /* the first bind address if available (or empty) */ |
| 69 | EXTCHK_HAPROXY_PROXY_PORT, /* the first bind port if available (or empty) */ |
| 70 | |
| 71 | /* Server specific environment variables */ |
| 72 | EXTCHK_HAPROXY_SERVER_NAME, /* the server name */ |
| 73 | EXTCHK_HAPROXY_SERVER_ID, /* the server id */ |
| 74 | EXTCHK_HAPROXY_SERVER_ADDR, /* the server address */ |
| 75 | EXTCHK_HAPROXY_SERVER_PORT, /* the server port if available (or empty) */ |
| 76 | EXTCHK_HAPROXY_SERVER_MAXCONN, /* the server max connections */ |
| 77 | EXTCHK_HAPROXY_SERVER_CURCONN, /* the current number of connections on the server */ |
| 78 | |
| 79 | EXTCHK_SIZE |
| 80 | }; |
| 81 | |
| 82 | const struct extcheck_env extcheck_envs[EXTCHK_SIZE] = { |
| 83 | [EXTCHK_PATH] = { "PATH", EXTCHK_SIZE_EVAL_INIT }, |
| 84 | [EXTCHK_HAPROXY_PROXY_NAME] = { "HAPROXY_PROXY_NAME", EXTCHK_SIZE_EVAL_INIT }, |
| 85 | [EXTCHK_HAPROXY_PROXY_ID] = { "HAPROXY_PROXY_ID", EXTCHK_SIZE_EVAL_INIT }, |
| 86 | [EXTCHK_HAPROXY_PROXY_ADDR] = { "HAPROXY_PROXY_ADDR", EXTCHK_SIZE_EVAL_INIT }, |
| 87 | [EXTCHK_HAPROXY_PROXY_PORT] = { "HAPROXY_PROXY_PORT", EXTCHK_SIZE_EVAL_INIT }, |
| 88 | [EXTCHK_HAPROXY_SERVER_NAME] = { "HAPROXY_SERVER_NAME", EXTCHK_SIZE_EVAL_INIT }, |
| 89 | [EXTCHK_HAPROXY_SERVER_ID] = { "HAPROXY_SERVER_ID", EXTCHK_SIZE_EVAL_INIT }, |
| 90 | [EXTCHK_HAPROXY_SERVER_ADDR] = { "HAPROXY_SERVER_ADDR", EXTCHK_SIZE_ADDR }, |
| 91 | [EXTCHK_HAPROXY_SERVER_PORT] = { "HAPROXY_SERVER_PORT", EXTCHK_SIZE_UINT }, |
| 92 | [EXTCHK_HAPROXY_SERVER_MAXCONN] = { "HAPROXY_SERVER_MAXCONN", EXTCHK_SIZE_EVAL_INIT }, |
| 93 | [EXTCHK_HAPROXY_SERVER_CURCONN] = { "HAPROXY_SERVER_CURCONN", EXTCHK_SIZE_ULONG }, |
| 94 | }; |
| 95 | |
| 96 | void block_sigchld(void) |
| 97 | { |
| 98 | sigset_t set; |
| 99 | sigemptyset(&set); |
| 100 | sigaddset(&set, SIGCHLD); |
| 101 | assert(ha_sigmask(SIG_BLOCK, &set, NULL) == 0); |
| 102 | } |
| 103 | |
| 104 | void unblock_sigchld(void) |
| 105 | { |
| 106 | sigset_t set; |
| 107 | sigemptyset(&set); |
| 108 | sigaddset(&set, SIGCHLD); |
| 109 | assert(ha_sigmask(SIG_UNBLOCK, &set, NULL) == 0); |
| 110 | } |
| 111 | |
| 112 | static struct pid_list *pid_list_add(pid_t pid, struct task *t) |
| 113 | { |
| 114 | struct pid_list *elem; |
| 115 | struct check *check = t->context; |
| 116 | |
| 117 | elem = pool_alloc(pool_head_pid_list); |
| 118 | if (!elem) |
| 119 | return NULL; |
| 120 | elem->pid = pid; |
| 121 | elem->t = t; |
| 122 | elem->exited = 0; |
| 123 | check->curpid = elem; |
| 124 | LIST_INIT(&elem->list); |
| 125 | |
| 126 | HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock); |
| 127 | LIST_ADD(&pid_list, &elem->list); |
| 128 | HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock); |
| 129 | |
| 130 | return elem; |
| 131 | } |
| 132 | |
| 133 | static void pid_list_del(struct pid_list *elem) |
| 134 | { |
| 135 | struct check *check; |
| 136 | |
| 137 | if (!elem) |
| 138 | return; |
| 139 | |
| 140 | HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock); |
| 141 | LIST_DEL(&elem->list); |
| 142 | HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock); |
| 143 | |
| 144 | if (!elem->exited) |
| 145 | kill(elem->pid, SIGTERM); |
| 146 | |
| 147 | check = elem->t->context; |
| 148 | check->curpid = NULL; |
| 149 | pool_free(pool_head_pid_list, elem); |
| 150 | } |
| 151 | |
| 152 | /* Called from inside SIGCHLD handler, SIGCHLD is blocked */ |
| 153 | static void pid_list_expire(pid_t pid, int status) |
| 154 | { |
| 155 | struct pid_list *elem; |
| 156 | |
| 157 | HA_SPIN_LOCK(PID_LIST_LOCK, &pid_list_lock); |
| 158 | list_for_each_entry(elem, &pid_list, list) { |
| 159 | if (elem->pid == pid) { |
| 160 | elem->t->expire = now_ms; |
| 161 | elem->status = status; |
| 162 | elem->exited = 1; |
| 163 | task_wakeup(elem->t, TASK_WOKEN_IO); |
| 164 | break; |
| 165 | } |
| 166 | } |
| 167 | HA_SPIN_UNLOCK(PID_LIST_LOCK, &pid_list_lock); |
| 168 | } |
| 169 | |
| 170 | static void sigchld_handler(struct sig_handler *sh) |
| 171 | { |
| 172 | pid_t pid; |
| 173 | int status; |
| 174 | |
| 175 | while ((pid = waitpid(0, &status, WNOHANG)) > 0) |
| 176 | pid_list_expire(pid, status); |
| 177 | } |
| 178 | |
| 179 | int init_pid_list(void) |
| 180 | { |
| 181 | if (pool_head_pid_list != NULL) |
| 182 | /* Nothing to do */ |
| 183 | return 0; |
| 184 | |
| 185 | if (!signal_register_fct(SIGCHLD, sigchld_handler, SIGCHLD)) { |
| 186 | ha_alert("Failed to set signal handler for external health checks: %s. Aborting.\n", |
| 187 | strerror(errno)); |
| 188 | return 1; |
| 189 | } |
| 190 | |
| 191 | pool_head_pid_list = create_pool("pid_list", sizeof(struct pid_list), MEM_F_SHARED); |
| 192 | if (pool_head_pid_list == NULL) { |
| 193 | ha_alert("Failed to allocate memory pool for external health checks: %s. Aborting.\n", |
| 194 | strerror(errno)); |
| 195 | return 1; |
| 196 | } |
| 197 | |
| 198 | return 0; |
| 199 | } |
| 200 | |
| 201 | /* helper macro to set an environment variable and jump to a specific label on failure. */ |
| 202 | #define EXTCHK_SETENV(check, envidx, value, fail) { if (extchk_setenv(check, envidx, value)) goto fail; } |
| 203 | |
| 204 | /* |
| 205 | * helper function to allocate enough memory to store an environment variable. |
| 206 | * It will also check that the environment variable is updatable, and silently |
| 207 | * fail if not. |
| 208 | */ |
| 209 | static int extchk_setenv(struct check *check, int idx, const char *value) |
| 210 | { |
| 211 | int len, ret; |
| 212 | char *envname; |
| 213 | int vmaxlen; |
| 214 | |
| 215 | if (idx < 0 || idx >= EXTCHK_SIZE) { |
| 216 | ha_alert("Illegal environment variable index %d. Aborting.\n", idx); |
| 217 | return 1; |
| 218 | } |
| 219 | |
| 220 | envname = extcheck_envs[idx].name; |
| 221 | vmaxlen = extcheck_envs[idx].vmaxlen; |
| 222 | |
| 223 | /* Check if the environment variable is already set, and silently reject |
| 224 | * the update if this one is not updatable. */ |
| 225 | if ((vmaxlen == EXTCHK_SIZE_EVAL_INIT) && (check->envp[idx])) |
| 226 | return 0; |
| 227 | |
| 228 | /* Instead of sending NOT_USED, sending an empty value is preferable */ |
| 229 | if (strcmp(value, "NOT_USED") == 0) { |
| 230 | value = ""; |
| 231 | } |
| 232 | |
| 233 | len = strlen(envname) + 1; |
| 234 | if (vmaxlen == EXTCHK_SIZE_EVAL_INIT) |
| 235 | len += strlen(value); |
| 236 | else |
| 237 | len += vmaxlen; |
| 238 | |
| 239 | if (!check->envp[idx]) |
| 240 | check->envp[idx] = malloc(len + 1); |
| 241 | |
| 242 | if (!check->envp[idx]) { |
| 243 | ha_alert("Failed to allocate memory for the environment variable '%s'. Aborting.\n", envname); |
| 244 | return 1; |
| 245 | } |
| 246 | ret = snprintf(check->envp[idx], len + 1, "%s=%s", envname, value); |
| 247 | if (ret < 0) { |
| 248 | ha_alert("Failed to store the environment variable '%s'. Reason : %s. Aborting.\n", envname, strerror(errno)); |
| 249 | return 1; |
| 250 | } |
| 251 | else if (ret > len) { |
| 252 | ha_alert("Environment variable '%s' was truncated. Aborting.\n", envname); |
| 253 | return 1; |
| 254 | } |
| 255 | return 0; |
| 256 | } |
| 257 | |
| 258 | int prepare_external_check(struct check *check) |
| 259 | { |
| 260 | struct server *s = check->server; |
| 261 | struct proxy *px = s->proxy; |
| 262 | struct listener *listener = NULL, *l; |
| 263 | int i; |
| 264 | const char *path = px->check_path ? px->check_path : DEF_CHECK_PATH; |
| 265 | char buf[256]; |
| 266 | |
| 267 | list_for_each_entry(l, &px->conf.listeners, by_fe) |
| 268 | /* Use the first INET, INET6 or UNIX listener */ |
Willy Tarreau | 3715906 | 2020-08-27 07:48:42 +0200 | [diff] [blame] | 269 | if (l->rx.addr.ss_family == AF_INET || |
| 270 | l->rx.addr.ss_family == AF_INET6 || |
| 271 | l->rx.addr.ss_family == AF_UNIX) { |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 272 | listener = l; |
| 273 | break; |
| 274 | } |
| 275 | |
| 276 | check->curpid = NULL; |
Tim Duesterhus | e52b6e5 | 2020-09-12 20:26:43 +0200 | [diff] [blame] | 277 | check->envp = calloc((EXTCHK_SIZE + 1), sizeof(*check->envp)); |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 278 | if (!check->envp) { |
| 279 | ha_alert("Failed to allocate memory for environment variables. Aborting\n"); |
| 280 | goto err; |
| 281 | } |
| 282 | |
Tim Duesterhus | e52b6e5 | 2020-09-12 20:26:43 +0200 | [diff] [blame] | 283 | check->argv = calloc(6, sizeof(*check->argv)); |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 284 | if (!check->argv) { |
| 285 | ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id); |
| 286 | goto err; |
| 287 | } |
| 288 | |
| 289 | check->argv[0] = px->check_command; |
| 290 | |
| 291 | if (!listener) { |
| 292 | check->argv[1] = strdup("NOT_USED"); |
| 293 | check->argv[2] = strdup("NOT_USED"); |
| 294 | } |
Willy Tarreau | 3715906 | 2020-08-27 07:48:42 +0200 | [diff] [blame] | 295 | else if (listener->rx.addr.ss_family == AF_INET || |
| 296 | listener->rx.addr.ss_family == AF_INET6) { |
| 297 | addr_to_str(&listener->rx.addr, buf, sizeof(buf)); |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 298 | check->argv[1] = strdup(buf); |
Willy Tarreau | 3715906 | 2020-08-27 07:48:42 +0200 | [diff] [blame] | 299 | port_to_str(&listener->rx.addr, buf, sizeof(buf)); |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 300 | check->argv[2] = strdup(buf); |
| 301 | } |
Willy Tarreau | 3715906 | 2020-08-27 07:48:42 +0200 | [diff] [blame] | 302 | else if (listener->rx.addr.ss_family == AF_UNIX) { |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 303 | const struct sockaddr_un *un; |
| 304 | |
Willy Tarreau | 3715906 | 2020-08-27 07:48:42 +0200 | [diff] [blame] | 305 | un = (struct sockaddr_un *)&listener->rx.addr; |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 306 | check->argv[1] = strdup(un->sun_path); |
| 307 | check->argv[2] = strdup("NOT_USED"); |
| 308 | } |
| 309 | else { |
| 310 | ha_alert("Starting [%s:%s] check: unsupported address family.\n", px->id, s->id); |
| 311 | goto err; |
| 312 | } |
| 313 | |
| 314 | if (!check->argv[1] || !check->argv[2]) { |
| 315 | ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id); |
| 316 | goto err; |
| 317 | } |
| 318 | |
| 319 | check->argv[3] = calloc(EXTCHK_SIZE_ADDR, sizeof(*check->argv[3])); |
| 320 | check->argv[4] = calloc(EXTCHK_SIZE_UINT, sizeof(*check->argv[4])); |
| 321 | if (!check->argv[3] || !check->argv[4]) { |
| 322 | ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id); |
| 323 | goto err; |
| 324 | } |
| 325 | |
| 326 | addr_to_str(&s->addr, check->argv[3], EXTCHK_SIZE_ADDR); |
| 327 | if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6) |
| 328 | snprintf(check->argv[4], EXTCHK_SIZE_UINT, "%u", s->svc_port); |
| 329 | |
| 330 | for (i = 0; i < 5; i++) { |
| 331 | if (!check->argv[i]) { |
| 332 | ha_alert("Starting [%s:%s] check: out of memory.\n", px->id, s->id); |
| 333 | goto err; |
| 334 | } |
| 335 | } |
| 336 | |
| 337 | EXTCHK_SETENV(check, EXTCHK_PATH, path, err); |
| 338 | /* Add proxy environment variables */ |
| 339 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_NAME, px->id, err); |
| 340 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ID, ultoa_r(px->uuid, buf, sizeof(buf)), err); |
| 341 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_ADDR, check->argv[1], err); |
| 342 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_PROXY_PORT, check->argv[2], err); |
| 343 | /* Add server environment variables */ |
| 344 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_NAME, s->id, err); |
| 345 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ID, ultoa_r(s->puid, buf, sizeof(buf)), err); |
| 346 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], err); |
| 347 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], err); |
| 348 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_MAXCONN, ultoa_r(s->maxconn, buf, sizeof(buf)), err); |
| 349 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), err); |
| 350 | |
| 351 | /* Ensure that we don't leave any hole in check->envp */ |
| 352 | for (i = 0; i < EXTCHK_SIZE; i++) |
| 353 | if (!check->envp[i]) |
| 354 | EXTCHK_SETENV(check, i, "", err); |
| 355 | |
| 356 | return 1; |
| 357 | err: |
| 358 | if (check->envp) { |
| 359 | for (i = 0; i < EXTCHK_SIZE; i++) |
| 360 | free(check->envp[i]); |
Willy Tarreau | 61cfdf4 | 2021-02-20 10:46:51 +0100 | [diff] [blame] | 361 | ha_free(&check->envp); |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 362 | } |
| 363 | |
| 364 | if (check->argv) { |
| 365 | for (i = 1; i < 5; i++) |
| 366 | free(check->argv[i]); |
Willy Tarreau | 61cfdf4 | 2021-02-20 10:46:51 +0100 | [diff] [blame] | 367 | ha_free(&check->argv); |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 368 | } |
| 369 | return 0; |
| 370 | } |
| 371 | |
| 372 | /* |
| 373 | * establish a server health-check that makes use of a process. |
| 374 | * |
| 375 | * It can return one of : |
| 376 | * - SF_ERR_NONE if everything's OK |
| 377 | * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...) |
| 378 | * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted. |
| 379 | * |
| 380 | * Blocks and then unblocks SIGCHLD |
| 381 | */ |
| 382 | static int connect_proc_chk(struct task *t) |
| 383 | { |
| 384 | char buf[256]; |
| 385 | struct check *check = t->context; |
| 386 | struct server *s = check->server; |
| 387 | struct proxy *px = s->proxy; |
| 388 | int status; |
| 389 | pid_t pid; |
| 390 | |
| 391 | status = SF_ERR_RESOURCE; |
| 392 | |
| 393 | block_sigchld(); |
| 394 | |
| 395 | pid = fork(); |
| 396 | if (pid < 0) { |
| 397 | ha_alert("Failed to fork process for external health check%s: %s. Aborting.\n", |
| 398 | (global.tune.options & GTUNE_INSECURE_FORK) ? |
| 399 | "" : " (likely caused by missing 'insecure-fork-wanted')", |
| 400 | strerror(errno)); |
| 401 | set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno)); |
| 402 | goto out; |
| 403 | } |
| 404 | if (pid == 0) { |
| 405 | /* Child */ |
| 406 | extern char **environ; |
| 407 | struct rlimit limit; |
| 408 | int fd; |
| 409 | |
| 410 | /* close all FDs. Keep stdin/stdout/stderr in verbose mode */ |
| 411 | fd = (global.mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_QUIET ? 0 : 3; |
| 412 | |
| 413 | my_closefrom(fd); |
| 414 | |
| 415 | /* restore the initial FD limits */ |
| 416 | limit.rlim_cur = rlim_fd_cur_at_boot; |
| 417 | limit.rlim_max = rlim_fd_max_at_boot; |
| 418 | if (setrlimit(RLIMIT_NOFILE, &limit) == -1) { |
| 419 | getrlimit(RLIMIT_NOFILE, &limit); |
| 420 | ha_warning("External check: failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n", |
| 421 | rlim_fd_cur_at_boot, rlim_fd_max_at_boot, |
| 422 | (unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max); |
| 423 | } |
| 424 | |
| 425 | environ = check->envp; |
| 426 | |
| 427 | /* Update some environment variables and command args: curconn, server addr and server port */ |
Willy Tarreau | b3250a2 | 2020-10-24 13:07:39 +0200 | [diff] [blame] | 428 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_CURCONN, ultoa_r(s->cur_sess, buf, sizeof(buf)), fail); |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 429 | |
| 430 | addr_to_str(&s->addr, check->argv[3], EXTCHK_SIZE_ADDR); |
Willy Tarreau | b3250a2 | 2020-10-24 13:07:39 +0200 | [diff] [blame] | 431 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_ADDR, check->argv[3], fail); |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 432 | |
| 433 | *check->argv[4] = 0; |
| 434 | if (s->addr.ss_family == AF_INET || s->addr.ss_family == AF_INET6) |
| 435 | snprintf(check->argv[4], EXTCHK_SIZE_UINT, "%u", s->svc_port); |
Willy Tarreau | b3250a2 | 2020-10-24 13:07:39 +0200 | [diff] [blame] | 436 | EXTCHK_SETENV(check, EXTCHK_HAPROXY_SERVER_PORT, check->argv[4], fail); |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 437 | |
| 438 | haproxy_unblock_signals(); |
| 439 | execvp(px->check_command, check->argv); |
| 440 | ha_alert("Failed to exec process for external health check: %s. Aborting.\n", |
| 441 | strerror(errno)); |
Willy Tarreau | b3250a2 | 2020-10-24 13:07:39 +0200 | [diff] [blame] | 442 | fail: |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 443 | exit(-1); |
| 444 | } |
| 445 | |
| 446 | /* Parent */ |
| 447 | if (check->result == CHK_RES_UNKNOWN) { |
| 448 | if (pid_list_add(pid, t) != NULL) { |
| 449 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
| 450 | |
| 451 | if (px->timeout.check && px->timeout.connect) { |
| 452 | int t_con = tick_add(now_ms, px->timeout.connect); |
| 453 | t->expire = tick_first(t->expire, t_con); |
| 454 | } |
| 455 | status = SF_ERR_NONE; |
| 456 | goto out; |
| 457 | } |
| 458 | else { |
| 459 | set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno)); |
| 460 | } |
| 461 | kill(pid, SIGTERM); /* process creation error */ |
| 462 | } |
| 463 | else |
| 464 | set_server_check_status(check, HCHK_STATUS_SOCKERR, strerror(errno)); |
| 465 | |
| 466 | out: |
| 467 | unblock_sigchld(); |
| 468 | return status; |
| 469 | } |
| 470 | |
| 471 | /* |
| 472 | * manages a server health-check that uses an external process. Returns |
| 473 | * the time the task accepts to wait, or TIME_ETERNITY for infinity. |
| 474 | * |
| 475 | * Please do NOT place any return statement in this function and only leave |
| 476 | * via the out_unlock label. |
| 477 | */ |
Willy Tarreau | 144f84a | 2021-03-02 16:09:26 +0100 | [diff] [blame] | 478 | struct task *process_chk_proc(struct task *t, void *context, unsigned int state) |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 479 | { |
| 480 | struct check *check = context; |
| 481 | struct server *s = check->server; |
| 482 | int rv; |
| 483 | int ret; |
| 484 | int expired = tick_is_expired(t->expire, now_ms); |
| 485 | |
| 486 | HA_SPIN_LOCK(SERVER_LOCK, &check->server->lock); |
| 487 | if (!(check->state & CHK_ST_INPROGRESS)) { |
| 488 | /* no check currently running */ |
| 489 | if (!expired) /* woke up too early */ |
| 490 | goto out_unlock; |
| 491 | |
| 492 | /* we don't send any health-checks when the proxy is |
| 493 | * stopped, the server should not be checked or the check |
| 494 | * is disabled. |
| 495 | */ |
| 496 | if (((check->state & (CHK_ST_ENABLED | CHK_ST_PAUSED)) != CHK_ST_ENABLED) || |
Willy Tarreau | c3914d4 | 2020-09-24 08:39:22 +0200 | [diff] [blame] | 497 | s->proxy->disabled) |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 498 | goto reschedule; |
| 499 | |
| 500 | /* we'll initiate a new check */ |
| 501 | set_server_check_status(check, HCHK_STATUS_START, NULL); |
| 502 | |
| 503 | check->state |= CHK_ST_INPROGRESS; |
| 504 | |
| 505 | ret = connect_proc_chk(t); |
| 506 | if (ret == SF_ERR_NONE) { |
| 507 | /* the process was forked, we allow up to min(inter, |
| 508 | * timeout.connect) for it to report its status, but |
| 509 | * only when timeout.check is set as it may be to short |
| 510 | * for a full check otherwise. |
| 511 | */ |
| 512 | t->expire = tick_add(now_ms, MS_TO_TICKS(check->inter)); |
| 513 | |
| 514 | if (s->proxy->timeout.check && s->proxy->timeout.connect) { |
| 515 | int t_con = tick_add(now_ms, s->proxy->timeout.connect); |
| 516 | t->expire = tick_first(t->expire, t_con); |
| 517 | } |
| 518 | task_set_affinity(t, tid_bit); |
| 519 | goto reschedule; |
| 520 | } |
| 521 | |
| 522 | /* here, we failed to start the check */ |
| 523 | |
| 524 | check->state &= ~CHK_ST_INPROGRESS; |
| 525 | check_notify_failure(check); |
| 526 | |
| 527 | /* we allow up to min(inter, timeout.connect) for a connection |
| 528 | * to establish but only when timeout.check is set |
| 529 | * as it may be to short for a full check otherwise |
| 530 | */ |
| 531 | while (tick_is_expired(t->expire, now_ms)) { |
| 532 | int t_con; |
| 533 | |
| 534 | t_con = tick_add(t->expire, s->proxy->timeout.connect); |
| 535 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
| 536 | |
| 537 | if (s->proxy->timeout.check) |
| 538 | t->expire = tick_first(t->expire, t_con); |
| 539 | } |
| 540 | } |
| 541 | else { |
| 542 | /* there was a test running. |
| 543 | * First, let's check whether there was an uncaught error, |
| 544 | * which can happen on connect timeout or error. |
| 545 | */ |
| 546 | if (check->result == CHK_RES_UNKNOWN) { |
| 547 | /* good connection is enough for pure TCP check */ |
| 548 | struct pid_list *elem = check->curpid; |
| 549 | int status = HCHK_STATUS_UNKNOWN; |
| 550 | |
| 551 | if (elem->exited) { |
| 552 | status = elem->status; /* Save in case the process exits between use below */ |
| 553 | if (!WIFEXITED(status)) |
| 554 | check->code = -1; |
| 555 | else |
| 556 | check->code = WEXITSTATUS(status); |
| 557 | if (!WIFEXITED(status) || WEXITSTATUS(status)) |
| 558 | status = HCHK_STATUS_PROCERR; |
| 559 | else |
| 560 | status = HCHK_STATUS_PROCOK; |
| 561 | } else if (expired) { |
| 562 | status = HCHK_STATUS_PROCTOUT; |
| 563 | ha_warning("kill %d\n", (int)elem->pid); |
| 564 | kill(elem->pid, SIGTERM); |
| 565 | } |
| 566 | set_server_check_status(check, status, NULL); |
| 567 | } |
| 568 | |
| 569 | if (check->result == CHK_RES_FAILED) { |
| 570 | /* a failure or timeout detected */ |
| 571 | check_notify_failure(check); |
| 572 | } |
| 573 | else if (check->result == CHK_RES_CONDPASS) { |
| 574 | /* check is OK but asks for stopping mode */ |
| 575 | check_notify_stopping(check); |
| 576 | } |
| 577 | else if (check->result == CHK_RES_PASSED) { |
| 578 | /* a success was detected */ |
| 579 | check_notify_success(check); |
| 580 | } |
| 581 | task_set_affinity(t, 1); |
| 582 | check->state &= ~CHK_ST_INPROGRESS; |
| 583 | |
| 584 | pid_list_del(check->curpid); |
| 585 | |
| 586 | rv = 0; |
| 587 | if (global.spread_checks > 0) { |
| 588 | rv = srv_getinter(check) * global.spread_checks / 100; |
| 589 | rv -= (int) (2 * rv * (ha_random32() / 4294967295.0)); |
| 590 | } |
| 591 | t->expire = tick_add(now_ms, MS_TO_TICKS(srv_getinter(check) + rv)); |
| 592 | } |
| 593 | |
| 594 | reschedule: |
| 595 | while (tick_is_expired(t->expire, now_ms)) |
| 596 | t->expire = tick_add(t->expire, MS_TO_TICKS(check->inter)); |
| 597 | |
| 598 | out_unlock: |
| 599 | HA_SPIN_UNLOCK(SERVER_LOCK, &check->server->lock); |
| 600 | return t; |
| 601 | } |
| 602 | |
| 603 | /* Parses the "external-check" proxy keyword */ |
| 604 | int proxy_parse_extcheck(char **args, int section, struct proxy *curpx, |
| 605 | struct proxy *defpx, const char *file, int line, |
| 606 | char **errmsg) |
| 607 | { |
| 608 | int cur_arg, ret = 0; |
| 609 | |
| 610 | cur_arg = 1; |
| 611 | if (!*(args[cur_arg])) { |
| 612 | memprintf(errmsg, "missing argument after '%s'.\n", args[0]); |
| 613 | goto error; |
| 614 | } |
| 615 | |
| 616 | if (strcmp(args[cur_arg], "command") == 0) { |
| 617 | if (too_many_args(2, args, errmsg, NULL)) |
| 618 | goto error; |
| 619 | if (!*(args[cur_arg+1])) { |
| 620 | memprintf(errmsg, "missing argument after '%s'.", args[cur_arg]); |
| 621 | goto error; |
| 622 | } |
| 623 | free(curpx->check_command); |
| 624 | curpx->check_command = strdup(args[cur_arg+1]); |
| 625 | } |
| 626 | else if (strcmp(args[cur_arg], "path") == 0) { |
| 627 | if (too_many_args(2, args, errmsg, NULL)) |
| 628 | goto error; |
| 629 | if (!*(args[cur_arg+1])) { |
| 630 | memprintf(errmsg, "missing argument after '%s'.", args[cur_arg]); |
| 631 | goto error; |
| 632 | } |
| 633 | free(curpx->check_path); |
| 634 | curpx->check_path = strdup(args[cur_arg+1]); |
| 635 | } |
| 636 | else { |
| 637 | memprintf(errmsg, "'%s' only supports 'command' and 'path'. but got '%s'.", |
| 638 | args[0], args[1]); |
| 639 | goto error; |
| 640 | } |
| 641 | |
| 642 | ret = (*errmsg != NULL); /* Handle warning */ |
| 643 | return ret; |
| 644 | |
| 645 | error: |
| 646 | return -1; |
| 647 | } |
| 648 | |
Willy Tarreau | 220fd70 | 2021-02-12 12:07:38 +0100 | [diff] [blame] | 649 | int proxy_parse_external_check_opt(char **args, int cur_arg, struct proxy *curpx, const struct proxy *defpx, |
Willy Tarreau | bcc6733 | 2020-06-05 15:31:31 +0200 | [diff] [blame] | 650 | const char *file, int line) |
| 651 | { |
| 652 | int err_code = 0; |
| 653 | |
| 654 | curpx->options2 &= ~PR_O2_CHK_ANY; |
| 655 | curpx->options2 |= PR_O2_EXT_CHK; |
| 656 | if (alertif_too_many_args_idx(0, 1, file, line, args, &err_code)) |
| 657 | goto out; |
| 658 | |
| 659 | out: |
| 660 | return err_code; |
| 661 | } |
| 662 | |
| 663 | static struct cfg_kw_list cfg_kws = {ILH, { |
| 664 | { CFG_LISTEN, "external-check", proxy_parse_extcheck }, |
| 665 | { 0, NULL, NULL }, |
| 666 | }}; |
| 667 | |
| 668 | INITCALL1(STG_REGISTER, cfg_register_keywords, &cfg_kws); |