hastd.c revision 217965
1/*- 2 * Copyright (c) 2009-2010 The FreeBSD Foundation 3 * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 4 * All rights reserved. 5 * 6 * This software was developed by Pawel Jakub Dawidek under sponsorship from 7 * the FreeBSD Foundation. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 */ 30 31#include <sys/cdefs.h> 32__FBSDID("$FreeBSD: head/sbin/hastd/hastd.c 217965 2011-01-27 19:24:07Z pjd $"); 33 34#include <sys/param.h> 35#include <sys/linker.h> 36#include <sys/module.h> 37#include <sys/wait.h> 38 39#include <assert.h> 40#include <err.h> 41#include <errno.h> 42#include <libutil.h> 43#include <signal.h> 44#include <stdbool.h> 45#include <stdio.h> 46#include <stdlib.h> 47#include <string.h> 48#include <sysexits.h> 49#include <unistd.h> 50 51#include <activemap.h> 52#include <pjdlog.h> 53 54#include "control.h" 55#include "event.h" 56#include "hast.h" 57#include "hast_proto.h" 58#include "hastd.h" 59#include "hooks.h" 60#include "subr.h" 61 62/* Path to configuration file. */ 63const char *cfgpath = HAST_CONFIG; 64/* Hastd configuration. */ 65static struct hastd_config *cfg; 66/* Was SIGINT or SIGTERM signal received? */ 67bool sigexit_received = false; 68/* PID file handle. */ 69struct pidfh *pfh; 70 71/* How often check for hooks running for too long. */ 72#define REPORT_INTERVAL 5 73 74static void 75usage(void) 76{ 77 78 errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]"); 79} 80 81static void 82g_gate_load(void) 83{ 84 85 if (modfind("g_gate") == -1) { 86 /* Not present in kernel, try loading it. */ 87 if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) { 88 if (errno != EEXIST) { 89 pjdlog_exit(EX_OSERR, 90 "Unable to load geom_gate module"); 91 } 92 } 93 } 94} 95 96static void 97child_exit_log(unsigned int pid, int status) 98{ 99 100 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 101 pjdlog_debug(1, "Worker process exited gracefully (pid=%u).", 102 pid); 103 } else if (WIFSIGNALED(status)) { 104 pjdlog_error("Worker process killed (pid=%u, signal=%d).", 105 pid, WTERMSIG(status)); 106 } else { 107 pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).", 108 pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1); 109 } 110} 111 112static void 113child_exit(void) 114{ 115 struct hast_resource *res; 116 int status; 117 pid_t pid; 118 119 while ((pid = wait3(&status, WNOHANG, NULL)) > 0) { 120 /* Find resource related to the process that just exited. */ 121 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 122 if (pid == res->hr_workerpid) 123 break; 124 } 125 if (res == NULL) { 126 /* 127 * This can happen when new connection arrives and we 128 * cancel child responsible for the old one or if this 129 * was hook which we executed. 130 */ 131 hook_check_one(pid, status); 132 continue; 133 } 134 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 135 role2str(res->hr_role)); 136 child_exit_log(pid, status); 137 child_cleanup(res); 138 if (res->hr_role == HAST_ROLE_PRIMARY) { 139 /* 140 * Restart child process if it was killed by signal 141 * or exited because of temporary problem. 142 */ 143 if (WIFSIGNALED(status) || 144 (WIFEXITED(status) && 145 WEXITSTATUS(status) == EX_TEMPFAIL)) { 146 sleep(1); 147 pjdlog_info("Restarting worker process."); 148 hastd_primary(res); 149 } else { 150 res->hr_role = HAST_ROLE_INIT; 151 pjdlog_info("Changing resource role back to %s.", 152 role2str(res->hr_role)); 153 } 154 } 155 pjdlog_prefix_set("%s", ""); 156 } 157} 158 159static bool 160resource_needs_restart(const struct hast_resource *res0, 161 const struct hast_resource *res1) 162{ 163 164 assert(strcmp(res0->hr_name, res1->hr_name) == 0); 165 166 if (strcmp(res0->hr_provname, res1->hr_provname) != 0) 167 return (true); 168 if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0) 169 return (true); 170 if (res0->hr_role == HAST_ROLE_INIT || 171 res0->hr_role == HAST_ROLE_SECONDARY) { 172 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 173 return (true); 174 if (res0->hr_replication != res1->hr_replication) 175 return (true); 176 if (res0->hr_timeout != res1->hr_timeout) 177 return (true); 178 if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 179 return (true); 180 } 181 return (false); 182} 183 184static bool 185resource_needs_reload(const struct hast_resource *res0, 186 const struct hast_resource *res1) 187{ 188 189 assert(strcmp(res0->hr_name, res1->hr_name) == 0); 190 assert(strcmp(res0->hr_provname, res1->hr_provname) == 0); 191 assert(strcmp(res0->hr_localpath, res1->hr_localpath) == 0); 192 193 if (res0->hr_role != HAST_ROLE_PRIMARY) 194 return (false); 195 196 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 197 return (true); 198 if (res0->hr_replication != res1->hr_replication) 199 return (true); 200 if (res0->hr_timeout != res1->hr_timeout) 201 return (true); 202 if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 203 return (true); 204 return (false); 205} 206 207static void 208resource_reload(const struct hast_resource *res) 209{ 210 struct nv *nvin, *nvout; 211 int error; 212 213 assert(res->hr_role == HAST_ROLE_PRIMARY); 214 215 nvout = nv_alloc(); 216 nv_add_uint8(nvout, HASTCTL_RELOAD, "cmd"); 217 nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr"); 218 nv_add_int32(nvout, (int32_t)res->hr_replication, "replication"); 219 nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout"); 220 nv_add_string(nvout, res->hr_exec, "exec"); 221 if (nv_error(nvout) != 0) { 222 nv_free(nvout); 223 pjdlog_error("Unable to allocate header for reload message."); 224 return; 225 } 226 if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) { 227 pjdlog_errno(LOG_ERR, "Unable to send reload message"); 228 nv_free(nvout); 229 return; 230 } 231 nv_free(nvout); 232 233 /* Receive response. */ 234 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) { 235 pjdlog_errno(LOG_ERR, "Unable to receive reload reply"); 236 return; 237 } 238 error = nv_get_int16(nvin, "error"); 239 nv_free(nvin); 240 if (error != 0) { 241 pjdlog_common(LOG_ERR, 0, error, "Reload failed"); 242 return; 243 } 244} 245 246static void 247hastd_reload(void) 248{ 249 struct hastd_config *newcfg; 250 struct hast_resource *nres, *cres, *tres; 251 uint8_t role; 252 253 pjdlog_info("Reloading configuration..."); 254 255 newcfg = yy_config_parse(cfgpath, false); 256 if (newcfg == NULL) 257 goto failed; 258 259 /* 260 * Check if control address has changed. 261 */ 262 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 263 if (proto_server(newcfg->hc_controladdr, 264 &newcfg->hc_controlconn) < 0) { 265 pjdlog_errno(LOG_ERR, 266 "Unable to listen on control address %s", 267 newcfg->hc_controladdr); 268 goto failed; 269 } 270 } 271 /* 272 * Check if listen address has changed. 273 */ 274 if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) { 275 if (proto_server(newcfg->hc_listenaddr, 276 &newcfg->hc_listenconn) < 0) { 277 pjdlog_errno(LOG_ERR, "Unable to listen on address %s", 278 newcfg->hc_listenaddr); 279 goto failed; 280 } 281 } 282 /* 283 * Only when both control and listen sockets are successfully 284 * initialized switch them to new configuration. 285 */ 286 if (newcfg->hc_controlconn != NULL) { 287 pjdlog_info("Control socket changed from %s to %s.", 288 cfg->hc_controladdr, newcfg->hc_controladdr); 289 proto_close(cfg->hc_controlconn); 290 cfg->hc_controlconn = newcfg->hc_controlconn; 291 newcfg->hc_controlconn = NULL; 292 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 293 sizeof(cfg->hc_controladdr)); 294 } 295 if (newcfg->hc_listenconn != NULL) { 296 pjdlog_info("Listen socket changed from %s to %s.", 297 cfg->hc_listenaddr, newcfg->hc_listenaddr); 298 proto_close(cfg->hc_listenconn); 299 cfg->hc_listenconn = newcfg->hc_listenconn; 300 newcfg->hc_listenconn = NULL; 301 strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr, 302 sizeof(cfg->hc_listenaddr)); 303 } 304 305 /* 306 * Stop and remove resources that were removed from the configuration. 307 */ 308 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 309 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 310 if (strcmp(cres->hr_name, nres->hr_name) == 0) 311 break; 312 } 313 if (nres == NULL) { 314 control_set_role(cres, HAST_ROLE_INIT); 315 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 316 pjdlog_info("Resource %s removed.", cres->hr_name); 317 free(cres); 318 } 319 } 320 /* 321 * Move new resources to the current configuration. 322 */ 323 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 324 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 325 if (strcmp(cres->hr_name, nres->hr_name) == 0) 326 break; 327 } 328 if (cres == NULL) { 329 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 330 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 331 pjdlog_info("Resource %s added.", nres->hr_name); 332 } 333 } 334 /* 335 * Deal with modified resources. 336 * Depending on what has changed exactly we might want to perform 337 * different actions. 338 * 339 * We do full resource restart in the following situations: 340 * Resource role is INIT or SECONDARY. 341 * Resource role is PRIMARY and path to local component or provider 342 * name has changed. 343 * In case of PRIMARY, the worker process will be killed and restarted, 344 * which also means removing /dev/hast/<name> provider and 345 * recreating it. 346 * 347 * We do just reload (send SIGHUP to worker process) if we act as 348 * PRIMARY, but only if remote address, replication mode, timeout or 349 * execution path has changed. For those, there is no need to restart 350 * worker process. 351 * If PRIMARY receives SIGHUP, it will reconnect if remote address or 352 * replication mode has changed or simply set new timeout if only 353 * timeout has changed. 354 */ 355 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 356 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 357 if (strcmp(cres->hr_name, nres->hr_name) == 0) 358 break; 359 } 360 assert(cres != NULL); 361 if (resource_needs_restart(cres, nres)) { 362 pjdlog_info("Resource %s configuration was modified, restarting it.", 363 cres->hr_name); 364 role = cres->hr_role; 365 control_set_role(cres, HAST_ROLE_INIT); 366 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 367 free(cres); 368 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 369 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 370 control_set_role(nres, role); 371 } else if (resource_needs_reload(cres, nres)) { 372 pjdlog_info("Resource %s configuration was modified, reloading it.", 373 cres->hr_name); 374 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 375 sizeof(cres->hr_remoteaddr)); 376 cres->hr_replication = nres->hr_replication; 377 cres->hr_timeout = nres->hr_timeout; 378 strlcpy(cres->hr_exec, nres->hr_exec, 379 sizeof(cres->hr_exec)); 380 if (cres->hr_workerpid != 0) 381 resource_reload(cres); 382 } 383 } 384 385 yy_config_free(newcfg); 386 pjdlog_info("Configuration reloaded successfully."); 387 return; 388failed: 389 if (newcfg != NULL) { 390 if (newcfg->hc_controlconn != NULL) 391 proto_close(newcfg->hc_controlconn); 392 if (newcfg->hc_listenconn != NULL) 393 proto_close(newcfg->hc_listenconn); 394 yy_config_free(newcfg); 395 } 396 pjdlog_warning("Configuration not reloaded."); 397} 398 399static void 400terminate_workers(void) 401{ 402 struct hast_resource *res; 403 404 pjdlog_info("Termination signal received, exiting."); 405 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 406 if (res->hr_workerpid == 0) 407 continue; 408 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 409 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 410 if (kill(res->hr_workerpid, SIGTERM) == 0) 411 continue; 412 pjdlog_errno(LOG_WARNING, 413 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 414 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 415 } 416} 417 418static void 419listen_accept(void) 420{ 421 struct hast_resource *res; 422 struct proto_conn *conn; 423 struct nv *nvin, *nvout, *nverr; 424 const char *resname; 425 const unsigned char *token; 426 char laddr[256], raddr[256]; 427 size_t size; 428 pid_t pid; 429 int status; 430 431 proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr)); 432 pjdlog_debug(1, "Accepting connection to %s.", laddr); 433 434 if (proto_accept(cfg->hc_listenconn, &conn) < 0) { 435 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 436 return; 437 } 438 439 proto_local_address(conn, laddr, sizeof(laddr)); 440 proto_remote_address(conn, raddr, sizeof(raddr)); 441 pjdlog_info("Connection from %s to %s.", raddr, laddr); 442 443 /* Error in setting timeout is not critical, but why should it fail? */ 444 if (proto_timeout(conn, HAST_TIMEOUT) < 0) 445 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 446 447 nvin = nvout = nverr = NULL; 448 449 /* 450 * Before receiving any data see if remote host have access to any 451 * resource. 452 */ 453 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 454 if (proto_address_match(conn, res->hr_remoteaddr)) 455 break; 456 } 457 if (res == NULL) { 458 pjdlog_error("Client %s isn't known.", raddr); 459 goto close; 460 } 461 /* Ok, remote host can access at least one resource. */ 462 463 if (hast_proto_recv_hdr(conn, &nvin) < 0) { 464 pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 465 raddr); 466 goto close; 467 } 468 469 resname = nv_get_string(nvin, "resource"); 470 if (resname == NULL) { 471 pjdlog_error("No 'resource' field in the header received from %s.", 472 raddr); 473 goto close; 474 } 475 pjdlog_debug(2, "%s: resource=%s", raddr, resname); 476 token = nv_get_uint8_array(nvin, &size, "token"); 477 /* 478 * NULL token means that this is first conection. 479 */ 480 if (token != NULL && size != sizeof(res->hr_token)) { 481 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 482 raddr, sizeof(res->hr_token), size); 483 goto close; 484 } 485 486 /* 487 * From now on we want to send errors to the remote node. 488 */ 489 nverr = nv_alloc(); 490 491 /* Find resource related to this connection. */ 492 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 493 if (strcmp(resname, res->hr_name) == 0) 494 break; 495 } 496 /* Have we found the resource? */ 497 if (res == NULL) { 498 pjdlog_error("No resource '%s' as requested by %s.", 499 resname, raddr); 500 nv_add_stringf(nverr, "errmsg", "Resource not configured."); 501 goto fail; 502 } 503 504 /* Now that we know resource name setup log prefix. */ 505 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 506 507 /* Does the remote host have access to this resource? */ 508 if (!proto_address_match(conn, res->hr_remoteaddr)) { 509 pjdlog_error("Client %s has no access to the resource.", raddr); 510 nv_add_stringf(nverr, "errmsg", "No access to the resource."); 511 goto fail; 512 } 513 /* Is the resource marked as secondary? */ 514 if (res->hr_role != HAST_ROLE_SECONDARY) { 515 pjdlog_error("We act as %s for the resource and not as %s as requested by %s.", 516 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 517 raddr); 518 nv_add_stringf(nverr, "errmsg", 519 "Remote node acts as %s for the resource and not as %s.", 520 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 521 goto fail; 522 } 523 /* Does token (if exists) match? */ 524 if (token != NULL && memcmp(token, res->hr_token, 525 sizeof(res->hr_token)) != 0) { 526 pjdlog_error("Token received from %s doesn't match.", raddr); 527 nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 528 goto fail; 529 } 530 /* 531 * If there is no token, but we have half-open connection 532 * (only remotein) or full connection (worker process is running) 533 * we have to cancel those and accept the new connection. 534 */ 535 if (token == NULL) { 536 assert(res->hr_remoteout == NULL); 537 pjdlog_debug(1, "Initial connection from %s.", raddr); 538 if (res->hr_workerpid != 0) { 539 assert(res->hr_remotein == NULL); 540 pjdlog_debug(1, 541 "Worker process exists (pid=%u), stopping it.", 542 (unsigned int)res->hr_workerpid); 543 /* Stop child process. */ 544 if (kill(res->hr_workerpid, SIGINT) < 0) { 545 pjdlog_errno(LOG_ERR, 546 "Unable to stop worker process (pid=%u)", 547 (unsigned int)res->hr_workerpid); 548 /* 549 * Other than logging the problem we 550 * ignore it - nothing smart to do. 551 */ 552 } 553 /* Wait for it to exit. */ 554 else if ((pid = waitpid(res->hr_workerpid, 555 &status, 0)) != res->hr_workerpid) { 556 /* We can only log the problem. */ 557 pjdlog_errno(LOG_ERR, 558 "Waiting for worker process (pid=%u) failed", 559 (unsigned int)res->hr_workerpid); 560 } else { 561 child_exit_log(res->hr_workerpid, status); 562 } 563 child_cleanup(res); 564 } else if (res->hr_remotein != NULL) { 565 char oaddr[256]; 566 567 proto_remote_address(res->hr_remotein, oaddr, 568 sizeof(oaddr)); 569 pjdlog_debug(1, 570 "Canceling half-open connection from %s on connection from %s.", 571 oaddr, raddr); 572 proto_close(res->hr_remotein); 573 res->hr_remotein = NULL; 574 } 575 } 576 577 /* 578 * Checks and cleanups are done. 579 */ 580 581 if (token == NULL) { 582 arc4random_buf(res->hr_token, sizeof(res->hr_token)); 583 nvout = nv_alloc(); 584 nv_add_uint8_array(nvout, res->hr_token, 585 sizeof(res->hr_token), "token"); 586 if (nv_error(nvout) != 0) { 587 pjdlog_common(LOG_ERR, 0, nv_error(nvout), 588 "Unable to prepare return header for %s", raddr); 589 nv_add_stringf(nverr, "errmsg", 590 "Remote node was unable to prepare return header: %s.", 591 strerror(nv_error(nvout))); 592 goto fail; 593 } 594 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) { 595 int error = errno; 596 597 pjdlog_errno(LOG_ERR, "Unable to send response to %s", 598 raddr); 599 nv_add_stringf(nverr, "errmsg", 600 "Remote node was unable to send response: %s.", 601 strerror(error)); 602 goto fail; 603 } 604 res->hr_remotein = conn; 605 pjdlog_debug(1, "Incoming connection from %s configured.", 606 raddr); 607 } else { 608 res->hr_remoteout = conn; 609 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 610 hastd_secondary(res, nvin); 611 } 612 nv_free(nvin); 613 nv_free(nvout); 614 nv_free(nverr); 615 pjdlog_prefix_set("%s", ""); 616 return; 617fail: 618 if (nv_error(nverr) != 0) { 619 pjdlog_common(LOG_ERR, 0, nv_error(nverr), 620 "Unable to prepare error header for %s", raddr); 621 goto close; 622 } 623 if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) { 624 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 625 goto close; 626 } 627close: 628 if (nvin != NULL) 629 nv_free(nvin); 630 if (nvout != NULL) 631 nv_free(nvout); 632 if (nverr != NULL) 633 nv_free(nverr); 634 proto_close(conn); 635 pjdlog_prefix_set("%s", ""); 636} 637 638static void 639main_loop(void) 640{ 641 struct hast_resource *res; 642 struct timeval seltimeout; 643 struct timespec sigtimeout; 644 int fd, maxfd, ret, signo; 645 sigset_t mask; 646 fd_set rfds; 647 648 seltimeout.tv_sec = REPORT_INTERVAL; 649 seltimeout.tv_usec = 0; 650 sigtimeout.tv_sec = 0; 651 sigtimeout.tv_nsec = 0; 652 653 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 654 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 655 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 656 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 657 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 658 659 pjdlog_info("Started successfully, running protocol version %d.", 660 HAST_PROTO_VERSION); 661 662 for (;;) { 663 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 664 switch (signo) { 665 case SIGINT: 666 case SIGTERM: 667 sigexit_received = true; 668 terminate_workers(); 669 exit(EX_OK); 670 break; 671 case SIGCHLD: 672 child_exit(); 673 break; 674 case SIGHUP: 675 hastd_reload(); 676 break; 677 default: 678 assert(!"invalid condition"); 679 } 680 } 681 682 /* Setup descriptors for select(2). */ 683 FD_ZERO(&rfds); 684 maxfd = fd = proto_descriptor(cfg->hc_controlconn); 685 assert(fd >= 0); 686 FD_SET(fd, &rfds); 687 fd = proto_descriptor(cfg->hc_listenconn); 688 assert(fd >= 0); 689 FD_SET(fd, &rfds); 690 maxfd = fd > maxfd ? fd : maxfd; 691 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 692 if (res->hr_event == NULL) 693 continue; 694 fd = proto_descriptor(res->hr_event); 695 assert(fd >= 0); 696 FD_SET(fd, &rfds); 697 maxfd = fd > maxfd ? fd : maxfd; 698 } 699 700 assert(maxfd + 1 <= (int)FD_SETSIZE); 701 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 702 if (ret == 0) 703 hook_check(); 704 else if (ret == -1) { 705 if (errno == EINTR) 706 continue; 707 KEEP_ERRNO((void)pidfile_remove(pfh)); 708 pjdlog_exit(EX_OSERR, "select() failed"); 709 } 710 711 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 712 control_handle(cfg); 713 if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds)) 714 listen_accept(); 715 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 716 if (res->hr_event == NULL) 717 continue; 718 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 719 if (event_recv(res) == 0) 720 continue; 721 /* The worker process exited? */ 722 proto_close(res->hr_event); 723 res->hr_event = NULL; 724 } 725 } 726 } 727} 728 729static void 730dummy_sighandler(int sig __unused) 731{ 732 /* Nothing to do. */ 733} 734 735int 736main(int argc, char *argv[]) 737{ 738 const char *pidfile; 739 pid_t otherpid; 740 bool foreground; 741 int debuglevel; 742 sigset_t mask; 743 744 foreground = false; 745 debuglevel = 0; 746 pidfile = HASTD_PIDFILE; 747 748 for (;;) { 749 int ch; 750 751 ch = getopt(argc, argv, "c:dFhP:"); 752 if (ch == -1) 753 break; 754 switch (ch) { 755 case 'c': 756 cfgpath = optarg; 757 break; 758 case 'd': 759 debuglevel++; 760 break; 761 case 'F': 762 foreground = true; 763 break; 764 case 'P': 765 pidfile = optarg; 766 break; 767 case 'h': 768 default: 769 usage(); 770 } 771 } 772 argc -= optind; 773 argv += optind; 774 775 pjdlog_init(PJDLOG_MODE_STD); 776 pjdlog_debug_set(debuglevel); 777 778 g_gate_load(); 779 780 pfh = pidfile_open(pidfile, 0600, &otherpid); 781 if (pfh == NULL) { 782 if (errno == EEXIST) { 783 pjdlog_exitx(EX_TEMPFAIL, 784 "Another hastd is already running, pid: %jd.", 785 (intmax_t)otherpid); 786 } 787 /* If we cannot create pidfile from other reasons, only warn. */ 788 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile"); 789 } 790 791 cfg = yy_config_parse(cfgpath, true); 792 assert(cfg != NULL); 793 794 /* 795 * Restore default actions for interesting signals in case parent 796 * process (like init(8)) decided to ignore some of them (like SIGHUP). 797 */ 798 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 799 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 800 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 801 /* 802 * Because SIGCHLD is ignored by default, setup dummy handler for it, 803 * so we can mask it. 804 */ 805 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 806 807 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 808 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 809 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 810 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 811 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 812 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 813 814 /* Listen on control address. */ 815 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) { 816 KEEP_ERRNO((void)pidfile_remove(pfh)); 817 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 818 cfg->hc_controladdr); 819 } 820 /* Listen for remote connections. */ 821 if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) { 822 KEEP_ERRNO((void)pidfile_remove(pfh)); 823 pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 824 cfg->hc_listenaddr); 825 } 826 827 if (!foreground) { 828 if (daemon(0, 0) < 0) { 829 KEEP_ERRNO((void)pidfile_remove(pfh)); 830 pjdlog_exit(EX_OSERR, "Unable to daemonize"); 831 } 832 833 /* Start logging to syslog. */ 834 pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 835 836 /* Write PID to a file. */ 837 if (pidfile_write(pfh) < 0) { 838 pjdlog_errno(LOG_WARNING, 839 "Unable to write PID to a file"); 840 } 841 } 842 843 hook_init(); 844 845 main_loop(); 846 847 exit(0); 848} 849