hastd.c revision 217729
1199536Smr/*- 2199536Smr * Copyright (c) 2009-2010 The FreeBSD Foundation 3199536Smr * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 4199536Smr * All rights reserved. 5199536Smr * 6199536Smr * This software was developed by Pawel Jakub Dawidek under sponsorship from 7199536Smr * the FreeBSD Foundation. 8199536Smr * 9199536Smr * Redistribution and use in source and binary forms, with or without 10199536Smr * modification, are permitted provided that the following conditions 11199536Smr * are met: 12199536Smr * 1. Redistributions of source code must retain the above copyright 13199536Smr * 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 217729 2011-01-22 22:31:55Z 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 208hastd_reload(void) 209{ 210 struct hastd_config *newcfg; 211 struct hast_resource *nres, *cres, *tres; 212 uint8_t role; 213 214 pjdlog_info("Reloading configuration..."); 215 216 newcfg = yy_config_parse(cfgpath, false); 217 if (newcfg == NULL) 218 goto failed; 219 220 /* 221 * Check if control address has changed. 222 */ 223 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 224 if (proto_server(newcfg->hc_controladdr, 225 &newcfg->hc_controlconn) < 0) { 226 pjdlog_errno(LOG_ERR, 227 "Unable to listen on control address %s", 228 newcfg->hc_controladdr); 229 goto failed; 230 } 231 } 232 /* 233 * Check if listen address has changed. 234 */ 235 if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) { 236 if (proto_server(newcfg->hc_listenaddr, 237 &newcfg->hc_listenconn) < 0) { 238 pjdlog_errno(LOG_ERR, "Unable to listen on address %s", 239 newcfg->hc_listenaddr); 240 goto failed; 241 } 242 } 243 /* 244 * Only when both control and listen sockets are successfully 245 * initialized switch them to new configuration. 246 */ 247 if (newcfg->hc_controlconn != NULL) { 248 pjdlog_info("Control socket changed from %s to %s.", 249 cfg->hc_controladdr, newcfg->hc_controladdr); 250 proto_close(cfg->hc_controlconn); 251 cfg->hc_controlconn = newcfg->hc_controlconn; 252 newcfg->hc_controlconn = NULL; 253 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 254 sizeof(cfg->hc_controladdr)); 255 } 256 if (newcfg->hc_listenconn != NULL) { 257 pjdlog_info("Listen socket changed from %s to %s.", 258 cfg->hc_listenaddr, newcfg->hc_listenaddr); 259 proto_close(cfg->hc_listenconn); 260 cfg->hc_listenconn = newcfg->hc_listenconn; 261 newcfg->hc_listenconn = NULL; 262 strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr, 263 sizeof(cfg->hc_listenaddr)); 264 } 265 266 /* 267 * Stop and remove resources that were removed from the configuration. 268 */ 269 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 270 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 271 if (strcmp(cres->hr_name, nres->hr_name) == 0) 272 break; 273 } 274 if (nres == NULL) { 275 control_set_role(cres, HAST_ROLE_INIT); 276 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 277 pjdlog_info("Resource %s removed.", cres->hr_name); 278 free(cres); 279 } 280 } 281 /* 282 * Move new resources to the current configuration. 283 */ 284 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 285 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 286 if (strcmp(cres->hr_name, nres->hr_name) == 0) 287 break; 288 } 289 if (cres == NULL) { 290 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 291 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 292 pjdlog_info("Resource %s added.", nres->hr_name); 293 } 294 } 295 /* 296 * Deal with modified resources. 297 * Depending on what has changed exactly we might want to perform 298 * different actions. 299 * 300 * We do full resource restart in the following situations: 301 * Resource role is INIT or SECONDARY. 302 * Resource role is PRIMARY and path to local component or provider 303 * name has changed. 304 * In case of PRIMARY, the worker process will be killed and restarted, 305 * which also means removing /dev/hast/<name> provider and 306 * recreating it. 307 * 308 * We do just reload (send SIGHUP to worker process) if we act as 309 * PRIMARY, but only if remote address, replication mode, timeout or 310 * execution path has changed. For those, there is no need to restart 311 * worker process. 312 * If PRIMARY receives SIGHUP, it will reconnect if remote address or 313 * replication mode has changed or simply set new timeout if only 314 * timeout has changed. 315 */ 316 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 317 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 318 if (strcmp(cres->hr_name, nres->hr_name) == 0) 319 break; 320 } 321 assert(cres != NULL); 322 if (resource_needs_restart(cres, nres)) { 323 pjdlog_info("Resource %s configuration was modified, restarting it.", 324 cres->hr_name); 325 role = cres->hr_role; 326 control_set_role(cres, HAST_ROLE_INIT); 327 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 328 free(cres); 329 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 330 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 331 control_set_role(nres, role); 332 } else if (resource_needs_reload(cres, nres)) { 333 pjdlog_info("Resource %s configuration was modified, reloading it.", 334 cres->hr_name); 335 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 336 sizeof(cres->hr_remoteaddr)); 337 cres->hr_replication = nres->hr_replication; 338 cres->hr_timeout = nres->hr_timeout; 339 strlcpy(cres->hr_exec, nres->hr_exec, 340 sizeof(cres->hr_exec)); 341 if (cres->hr_workerpid != 0) { 342 if (kill(cres->hr_workerpid, SIGHUP) < 0) { 343 pjdlog_errno(LOG_WARNING, 344 "Unable to send SIGHUP to worker process %u", 345 (unsigned int)cres->hr_workerpid); 346 } 347 } 348 } 349 } 350 351 yy_config_free(newcfg); 352 pjdlog_info("Configuration reloaded successfully."); 353 return; 354failed: 355 if (newcfg != NULL) { 356 if (newcfg->hc_controlconn != NULL) 357 proto_close(newcfg->hc_controlconn); 358 if (newcfg->hc_listenconn != NULL) 359 proto_close(newcfg->hc_listenconn); 360 yy_config_free(newcfg); 361 } 362 pjdlog_warning("Configuration not reloaded."); 363} 364 365static void 366terminate_workers(void) 367{ 368 struct hast_resource *res; 369 370 pjdlog_info("Termination signal received, exiting."); 371 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 372 if (res->hr_workerpid == 0) 373 continue; 374 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 375 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 376 if (kill(res->hr_workerpid, SIGTERM) == 0) 377 continue; 378 pjdlog_errno(LOG_WARNING, 379 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 380 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 381 } 382} 383 384static void 385listen_accept(void) 386{ 387 struct hast_resource *res; 388 struct proto_conn *conn; 389 struct nv *nvin, *nvout, *nverr; 390 const char *resname; 391 const unsigned char *token; 392 char laddr[256], raddr[256]; 393 size_t size; 394 pid_t pid; 395 int status; 396 397 proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr)); 398 pjdlog_debug(1, "Accepting connection to %s.", laddr); 399 400 if (proto_accept(cfg->hc_listenconn, &conn) < 0) { 401 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 402 return; 403 } 404 405 proto_local_address(conn, laddr, sizeof(laddr)); 406 proto_remote_address(conn, raddr, sizeof(raddr)); 407 pjdlog_info("Connection from %s to %s.", raddr, laddr); 408 409 /* Error in setting timeout is not critical, but why should it fail? */ 410 if (proto_timeout(conn, HAST_TIMEOUT) < 0) 411 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 412 413 nvin = nvout = nverr = NULL; 414 415 /* 416 * Before receiving any data see if remote host have access to any 417 * resource. 418 */ 419 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 420 if (proto_address_match(conn, res->hr_remoteaddr)) 421 break; 422 } 423 if (res == NULL) { 424 pjdlog_error("Client %s isn't known.", raddr); 425 goto close; 426 } 427 /* Ok, remote host can access at least one resource. */ 428 429 if (hast_proto_recv_hdr(conn, &nvin) < 0) { 430 pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 431 raddr); 432 goto close; 433 } 434 435 resname = nv_get_string(nvin, "resource"); 436 if (resname == NULL) { 437 pjdlog_error("No 'resource' field in the header received from %s.", 438 raddr); 439 goto close; 440 } 441 pjdlog_debug(2, "%s: resource=%s", raddr, resname); 442 token = nv_get_uint8_array(nvin, &size, "token"); 443 /* 444 * NULL token means that this is first conection. 445 */ 446 if (token != NULL && size != sizeof(res->hr_token)) { 447 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 448 raddr, sizeof(res->hr_token), size); 449 goto close; 450 } 451 452 /* 453 * From now on we want to send errors to the remote node. 454 */ 455 nverr = nv_alloc(); 456 457 /* Find resource related to this connection. */ 458 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 459 if (strcmp(resname, res->hr_name) == 0) 460 break; 461 } 462 /* Have we found the resource? */ 463 if (res == NULL) { 464 pjdlog_error("No resource '%s' as requested by %s.", 465 resname, raddr); 466 nv_add_stringf(nverr, "errmsg", "Resource not configured."); 467 goto fail; 468 } 469 470 /* Now that we know resource name setup log prefix. */ 471 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 472 473 /* Does the remote host have access to this resource? */ 474 if (!proto_address_match(conn, res->hr_remoteaddr)) { 475 pjdlog_error("Client %s has no access to the resource.", raddr); 476 nv_add_stringf(nverr, "errmsg", "No access to the resource."); 477 goto fail; 478 } 479 /* Is the resource marked as secondary? */ 480 if (res->hr_role != HAST_ROLE_SECONDARY) { 481 pjdlog_error("We act as %s for the resource and not as %s as requested by %s.", 482 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 483 raddr); 484 nv_add_stringf(nverr, "errmsg", 485 "Remote node acts as %s for the resource and not as %s.", 486 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 487 goto fail; 488 } 489 /* Does token (if exists) match? */ 490 if (token != NULL && memcmp(token, res->hr_token, 491 sizeof(res->hr_token)) != 0) { 492 pjdlog_error("Token received from %s doesn't match.", raddr); 493 nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 494 goto fail; 495 } 496 /* 497 * If there is no token, but we have half-open connection 498 * (only remotein) or full connection (worker process is running) 499 * we have to cancel those and accept the new connection. 500 */ 501 if (token == NULL) { 502 assert(res->hr_remoteout == NULL); 503 pjdlog_debug(1, "Initial connection from %s.", raddr); 504 if (res->hr_workerpid != 0) { 505 assert(res->hr_remotein == NULL); 506 pjdlog_debug(1, 507 "Worker process exists (pid=%u), stopping it.", 508 (unsigned int)res->hr_workerpid); 509 /* Stop child process. */ 510 if (kill(res->hr_workerpid, SIGINT) < 0) { 511 pjdlog_errno(LOG_ERR, 512 "Unable to stop worker process (pid=%u)", 513 (unsigned int)res->hr_workerpid); 514 /* 515 * Other than logging the problem we 516 * ignore it - nothing smart to do. 517 */ 518 } 519 /* Wait for it to exit. */ 520 else if ((pid = waitpid(res->hr_workerpid, 521 &status, 0)) != res->hr_workerpid) { 522 /* We can only log the problem. */ 523 pjdlog_errno(LOG_ERR, 524 "Waiting for worker process (pid=%u) failed", 525 (unsigned int)res->hr_workerpid); 526 } else { 527 child_exit_log(res->hr_workerpid, status); 528 } 529 child_cleanup(res); 530 } else if (res->hr_remotein != NULL) { 531 char oaddr[256]; 532 533 proto_remote_address(res->hr_remotein, oaddr, 534 sizeof(oaddr)); 535 pjdlog_debug(1, 536 "Canceling half-open connection from %s on connection from %s.", 537 oaddr, raddr); 538 proto_close(res->hr_remotein); 539 res->hr_remotein = NULL; 540 } 541 } 542 543 /* 544 * Checks and cleanups are done. 545 */ 546 547 if (token == NULL) { 548 arc4random_buf(res->hr_token, sizeof(res->hr_token)); 549 nvout = nv_alloc(); 550 nv_add_uint8_array(nvout, res->hr_token, 551 sizeof(res->hr_token), "token"); 552 if (nv_error(nvout) != 0) { 553 pjdlog_common(LOG_ERR, 0, nv_error(nvout), 554 "Unable to prepare return header for %s", raddr); 555 nv_add_stringf(nverr, "errmsg", 556 "Remote node was unable to prepare return header: %s.", 557 strerror(nv_error(nvout))); 558 goto fail; 559 } 560 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) { 561 int error = errno; 562 563 pjdlog_errno(LOG_ERR, "Unable to send response to %s", 564 raddr); 565 nv_add_stringf(nverr, "errmsg", 566 "Remote node was unable to send response: %s.", 567 strerror(error)); 568 goto fail; 569 } 570 res->hr_remotein = conn; 571 pjdlog_debug(1, "Incoming connection from %s configured.", 572 raddr); 573 } else { 574 res->hr_remoteout = conn; 575 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 576 hastd_secondary(res, nvin); 577 } 578 nv_free(nvin); 579 nv_free(nvout); 580 nv_free(nverr); 581 pjdlog_prefix_set("%s", ""); 582 return; 583fail: 584 if (nv_error(nverr) != 0) { 585 pjdlog_common(LOG_ERR, 0, nv_error(nverr), 586 "Unable to prepare error header for %s", raddr); 587 goto close; 588 } 589 if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) { 590 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 591 goto close; 592 } 593close: 594 if (nvin != NULL) 595 nv_free(nvin); 596 if (nvout != NULL) 597 nv_free(nvout); 598 if (nverr != NULL) 599 nv_free(nverr); 600 proto_close(conn); 601 pjdlog_prefix_set("%s", ""); 602} 603 604static void 605main_loop(void) 606{ 607 struct hast_resource *res; 608 struct timeval seltimeout; 609 struct timespec sigtimeout; 610 int fd, maxfd, ret, signo; 611 sigset_t mask; 612 fd_set rfds; 613 614 seltimeout.tv_sec = REPORT_INTERVAL; 615 seltimeout.tv_usec = 0; 616 sigtimeout.tv_sec = 0; 617 sigtimeout.tv_nsec = 0; 618 619 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 620 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 621 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 622 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 623 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 624 625 pjdlog_info("Started successfully, running protocol version %d.", 626 HAST_PROTO_VERSION); 627 628 for (;;) { 629 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 630 switch (signo) { 631 case SIGINT: 632 case SIGTERM: 633 sigexit_received = true; 634 terminate_workers(); 635 exit(EX_OK); 636 break; 637 case SIGCHLD: 638 child_exit(); 639 break; 640 case SIGHUP: 641 hastd_reload(); 642 break; 643 default: 644 assert(!"invalid condition"); 645 } 646 } 647 648 /* Setup descriptors for select(2). */ 649 FD_ZERO(&rfds); 650 maxfd = fd = proto_descriptor(cfg->hc_controlconn); 651 assert(fd >= 0); 652 FD_SET(fd, &rfds); 653 fd = proto_descriptor(cfg->hc_listenconn); 654 assert(fd >= 0); 655 FD_SET(fd, &rfds); 656 maxfd = fd > maxfd ? fd : maxfd; 657 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 658 if (res->hr_event == NULL) 659 continue; 660 fd = proto_descriptor(res->hr_event); 661 assert(fd >= 0); 662 FD_SET(fd, &rfds); 663 maxfd = fd > maxfd ? fd : maxfd; 664 } 665 666 assert(maxfd + 1 <= (int)FD_SETSIZE); 667 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 668 if (ret == 0) 669 hook_check(); 670 else if (ret == -1) { 671 if (errno == EINTR) 672 continue; 673 KEEP_ERRNO((void)pidfile_remove(pfh)); 674 pjdlog_exit(EX_OSERR, "select() failed"); 675 } 676 677 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 678 control_handle(cfg); 679 if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds)) 680 listen_accept(); 681 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 682 if (res->hr_event == NULL) 683 continue; 684 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 685 if (event_recv(res) == 0) 686 continue; 687 /* The worker process exited? */ 688 proto_close(res->hr_event); 689 res->hr_event = NULL; 690 } 691 } 692 } 693} 694 695static void 696dummy_sighandler(int sig __unused) 697{ 698 /* Nothing to do. */ 699} 700 701int 702main(int argc, char *argv[]) 703{ 704 const char *pidfile; 705 pid_t otherpid; 706 bool foreground; 707 int debuglevel; 708 sigset_t mask; 709 710 foreground = false; 711 debuglevel = 0; 712 pidfile = HASTD_PIDFILE; 713 714 for (;;) { 715 int ch; 716 717 ch = getopt(argc, argv, "c:dFhP:"); 718 if (ch == -1) 719 break; 720 switch (ch) { 721 case 'c': 722 cfgpath = optarg; 723 break; 724 case 'd': 725 debuglevel++; 726 break; 727 case 'F': 728 foreground = true; 729 break; 730 case 'P': 731 pidfile = optarg; 732 break; 733 case 'h': 734 default: 735 usage(); 736 } 737 } 738 argc -= optind; 739 argv += optind; 740 741 pjdlog_debug_set(debuglevel); 742 743 g_gate_load(); 744 745 pfh = pidfile_open(pidfile, 0600, &otherpid); 746 if (pfh == NULL) { 747 if (errno == EEXIST) { 748 pjdlog_exitx(EX_TEMPFAIL, 749 "Another hastd is already running, pid: %jd.", 750 (intmax_t)otherpid); 751 } 752 /* If we cannot create pidfile from other reasons, only warn. */ 753 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile"); 754 } 755 756 cfg = yy_config_parse(cfgpath, true); 757 assert(cfg != NULL); 758 759 /* 760 * Restore default actions for interesting signals in case parent 761 * process (like init(8)) decided to ignore some of them (like SIGHUP). 762 */ 763 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 764 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 765 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 766 /* 767 * Because SIGCHLD is ignored by default, setup dummy handler for it, 768 * so we can mask it. 769 */ 770 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 771 772 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 773 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 774 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 775 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 776 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 777 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 778 779 /* Listen on control address. */ 780 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) { 781 KEEP_ERRNO((void)pidfile_remove(pfh)); 782 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 783 cfg->hc_controladdr); 784 } 785 /* Listen for remote connections. */ 786 if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) { 787 KEEP_ERRNO((void)pidfile_remove(pfh)); 788 pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 789 cfg->hc_listenaddr); 790 } 791 792 if (!foreground) { 793 if (daemon(0, 0) < 0) { 794 KEEP_ERRNO((void)pidfile_remove(pfh)); 795 pjdlog_exit(EX_OSERR, "Unable to daemonize"); 796 } 797 798 /* Start logging to syslog. */ 799 pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 800 801 /* Write PID to a file. */ 802 if (pidfile_write(pfh) < 0) { 803 pjdlog_errno(LOG_WARNING, 804 "Unable to write PID to a file"); 805 } 806 } 807 808 hook_init(); 809 810 main_loop(); 811 812 exit(0); 813} 814