1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2009-2010 The FreeBSD Foundation 5 * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net> 6 * All rights reserved. 7 * 8 * This software was developed by Pawel Jakub Dawidek under sponsorship from 9 * the FreeBSD Foundation. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 */ 32 33 #include <sys/cdefs.h> 34 __FBSDID("$FreeBSD$"); 35 36 #include <sys/param.h> 37 #include <sys/linker.h> 38 #include <sys/module.h> 39 #include <sys/stat.h> 40 #include <sys/wait.h> 41 42 #include <err.h> 43 #include <errno.h> 44 #include <libutil.h> 45 #include <signal.h> 46 #include <stdbool.h> 47 #include <stdio.h> 48 #include <stdlib.h> 49 #include <string.h> 50 #include <sysexits.h> 51 #include <time.h> 52 #include <unistd.h> 53 54 #include <activemap.h> 55 #include <pjdlog.h> 56 57 #include "control.h" 58 #include "event.h" 59 #include "hast.h" 60 #include "hast_proto.h" 61 #include "hastd.h" 62 #include "hooks.h" 63 #include "subr.h" 64 65 /* Path to configuration file. */ 66 const char *cfgpath = HAST_CONFIG; 67 /* Hastd configuration. */ 68 static struct hastd_config *cfg; 69 /* Was SIGINT or SIGTERM signal received? */ 70 bool sigexit_received = false; 71 /* Path to pidfile. */ 72 static const char *pidfile; 73 /* Pidfile handle. */ 74 struct pidfh *pfh; 75 /* Do we run in foreground? */ 76 static bool foreground; 77 78 /* How often check for hooks running for too long. */ 79 #define REPORT_INTERVAL 5 80 81 static void 82 usage(void) 83 { 84 85 errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]"); 86 } 87 88 static void 89 g_gate_load(void) 90 { 91 92 if (modfind("g_gate") == -1) { 93 /* Not present in kernel, try loading it. */ 94 if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) { 95 if (errno != EEXIST) { 96 pjdlog_exit(EX_OSERR, 97 "Unable to load geom_gate module"); 98 } 99 } 100 } 101 } 102 103 void 104 descriptors_cleanup(struct hast_resource *res) 105 { 106 struct hast_resource *tres, *tmres; 107 struct hastd_listen *lst; 108 109 TAILQ_FOREACH_SAFE(tres, &cfg->hc_resources, hr_next, tmres) { 110 if (tres == res) { 111 PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY || 112 (res->hr_remotein == NULL && 113 res->hr_remoteout == NULL)); 114 continue; 115 } 116 if (tres->hr_remotein != NULL) 117 proto_close(tres->hr_remotein); 118 if (tres->hr_remoteout != NULL) 119 proto_close(tres->hr_remoteout); 120 if (tres->hr_ctrl != NULL) 121 proto_close(tres->hr_ctrl); 122 if (tres->hr_event != NULL) 123 proto_close(tres->hr_event); 124 if (tres->hr_conn != NULL) 125 proto_close(tres->hr_conn); 126 TAILQ_REMOVE(&cfg->hc_resources, tres, hr_next); 127 free(tres); 128 } 129 if (cfg->hc_controlin != NULL) 130 proto_close(cfg->hc_controlin); 131 proto_close(cfg->hc_controlconn); 132 while ((lst = TAILQ_FIRST(&cfg->hc_listen)) != NULL) { 133 TAILQ_REMOVE(&cfg->hc_listen, lst, hl_next); 134 if (lst->hl_conn != NULL) 135 proto_close(lst->hl_conn); 136 free(lst); 137 } 138 (void)pidfile_close(pfh); 139 hook_fini(); 140 pjdlog_fini(); 141 } 142 143 static const char * 144 dtype2str(mode_t mode) 145 { 146 147 if (S_ISBLK(mode)) 148 return ("block device"); 149 else if (S_ISCHR(mode)) 150 return ("character device"); 151 else if (S_ISDIR(mode)) 152 return ("directory"); 153 else if (S_ISFIFO(mode)) 154 return ("pipe or FIFO"); 155 else if (S_ISLNK(mode)) 156 return ("symbolic link"); 157 else if (S_ISREG(mode)) 158 return ("regular file"); 159 else if (S_ISSOCK(mode)) 160 return ("socket"); 161 else if (S_ISWHT(mode)) 162 return ("whiteout"); 163 else 164 return ("unknown"); 165 } 166 167 void 168 descriptors_assert(const struct hast_resource *res, int pjdlogmode) 169 { 170 char msg[256]; 171 struct stat sb; 172 long maxfd; 173 bool isopen; 174 mode_t mode; 175 int fd; 176 177 /* 178 * At this point descriptor to syslog socket is closed, so if we want 179 * to log assertion message, we have to first store it in 'msg' local 180 * buffer and then open syslog socket and log it. 181 */ 182 msg[0] = '\0'; 183 184 maxfd = sysconf(_SC_OPEN_MAX); 185 if (maxfd == -1) { 186 pjdlog_init(pjdlogmode); 187 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 188 role2str(res->hr_role)); 189 pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed"); 190 pjdlog_fini(); 191 maxfd = 16384; 192 } 193 for (fd = 0; fd <= maxfd; fd++) { 194 if (fstat(fd, &sb) == 0) { 195 isopen = true; 196 mode = sb.st_mode; 197 } else if (errno == EBADF) { 198 isopen = false; 199 mode = 0; 200 } else { 201 (void)snprintf(msg, sizeof(msg), 202 "Unable to fstat descriptor %d: %s", fd, 203 strerror(errno)); 204 break; 205 } 206 if (fd == STDIN_FILENO || fd == STDOUT_FILENO || 207 fd == STDERR_FILENO) { 208 if (!isopen) { 209 (void)snprintf(msg, sizeof(msg), 210 "Descriptor %d (%s) is closed, but should be open.", 211 fd, (fd == STDIN_FILENO ? "stdin" : 212 (fd == STDOUT_FILENO ? "stdout" : "stderr"))); 213 break; 214 } 215 } else if (fd == proto_descriptor(res->hr_event)) { 216 if (!isopen) { 217 (void)snprintf(msg, sizeof(msg), 218 "Descriptor %d (event) is closed, but should be open.", 219 fd); 220 break; 221 } 222 if (!S_ISSOCK(mode)) { 223 (void)snprintf(msg, sizeof(msg), 224 "Descriptor %d (event) is %s, but should be %s.", 225 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 226 break; 227 } 228 } else if (fd == proto_descriptor(res->hr_ctrl)) { 229 if (!isopen) { 230 (void)snprintf(msg, sizeof(msg), 231 "Descriptor %d (ctrl) is closed, but should be open.", 232 fd); 233 break; 234 } 235 if (!S_ISSOCK(mode)) { 236 (void)snprintf(msg, sizeof(msg), 237 "Descriptor %d (ctrl) is %s, but should be %s.", 238 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 239 break; 240 } 241 } else if (res->hr_role == HAST_ROLE_PRIMARY && 242 fd == proto_descriptor(res->hr_conn)) { 243 if (!isopen) { 244 (void)snprintf(msg, sizeof(msg), 245 "Descriptor %d (conn) is closed, but should be open.", 246 fd); 247 break; 248 } 249 if (!S_ISSOCK(mode)) { 250 (void)snprintf(msg, sizeof(msg), 251 "Descriptor %d (conn) is %s, but should be %s.", 252 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 253 break; 254 } 255 } else if (res->hr_role == HAST_ROLE_SECONDARY && 256 res->hr_conn != NULL && 257 fd == proto_descriptor(res->hr_conn)) { 258 if (isopen) { 259 (void)snprintf(msg, sizeof(msg), 260 "Descriptor %d (conn) is open, but should be closed.", 261 fd); 262 break; 263 } 264 } else if (res->hr_role == HAST_ROLE_SECONDARY && 265 fd == proto_descriptor(res->hr_remotein)) { 266 if (!isopen) { 267 (void)snprintf(msg, sizeof(msg), 268 "Descriptor %d (remote in) is closed, but should be open.", 269 fd); 270 break; 271 } 272 if (!S_ISSOCK(mode)) { 273 (void)snprintf(msg, sizeof(msg), 274 "Descriptor %d (remote in) is %s, but should be %s.", 275 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 276 break; 277 } 278 } else if (res->hr_role == HAST_ROLE_SECONDARY && 279 fd == proto_descriptor(res->hr_remoteout)) { 280 if (!isopen) { 281 (void)snprintf(msg, sizeof(msg), 282 "Descriptor %d (remote out) is closed, but should be open.", 283 fd); 284 break; 285 } 286 if (!S_ISSOCK(mode)) { 287 (void)snprintf(msg, sizeof(msg), 288 "Descriptor %d (remote out) is %s, but should be %s.", 289 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 290 break; 291 } 292 } else { 293 if (isopen) { 294 (void)snprintf(msg, sizeof(msg), 295 "Descriptor %d is open (%s), but should be closed.", 296 fd, dtype2str(mode)); 297 break; 298 } 299 } 300 } 301 if (msg[0] != '\0') { 302 pjdlog_init(pjdlogmode); 303 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 304 role2str(res->hr_role)); 305 PJDLOG_ABORT("%s", msg); 306 } 307 } 308 309 static void 310 child_exit_log(unsigned int pid, int status) 311 { 312 313 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 314 pjdlog_debug(1, "Worker process exited gracefully (pid=%u).", 315 pid); 316 } else if (WIFSIGNALED(status)) { 317 pjdlog_error("Worker process killed (pid=%u, signal=%d).", 318 pid, WTERMSIG(status)); 319 } else { 320 pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).", 321 pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1); 322 } 323 } 324 325 static void 326 child_exit(void) 327 { 328 struct hast_resource *res; 329 int status; 330 pid_t pid; 331 332 while ((pid = wait3(&status, WNOHANG, NULL)) > 0) { 333 /* Find resource related to the process that just exited. */ 334 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 335 if (pid == res->hr_workerpid) 336 break; 337 } 338 if (res == NULL) { 339 /* 340 * This can happen when new connection arrives and we 341 * cancel child responsible for the old one or if this 342 * was hook which we executed. 343 */ 344 hook_check_one(pid, status); 345 continue; 346 } 347 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 348 role2str(res->hr_role)); 349 child_exit_log(pid, status); 350 child_cleanup(res); 351 if (res->hr_role == HAST_ROLE_PRIMARY) { 352 /* 353 * Restart child process if it was killed by signal 354 * or exited because of temporary problem. 355 */ 356 if (WIFSIGNALED(status) || 357 (WIFEXITED(status) && 358 WEXITSTATUS(status) == EX_TEMPFAIL)) { 359 sleep(1); 360 pjdlog_info("Restarting worker process."); 361 hastd_primary(res); 362 } else { 363 res->hr_role = HAST_ROLE_INIT; 364 pjdlog_info("Changing resource role back to %s.", 365 role2str(res->hr_role)); 366 } 367 } 368 pjdlog_prefix_set("%s", ""); 369 } 370 } 371 372 static bool 373 resource_needs_restart(const struct hast_resource *res0, 374 const struct hast_resource *res1) 375 { 376 377 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 378 379 if (strcmp(res0->hr_provname, res1->hr_provname) != 0) 380 return (true); 381 if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0) 382 return (true); 383 if (res0->hr_role == HAST_ROLE_INIT || 384 res0->hr_role == HAST_ROLE_SECONDARY) { 385 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 386 return (true); 387 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0) 388 return (true); 389 if (res0->hr_replication != res1->hr_replication) 390 return (true); 391 if (res0->hr_checksum != res1->hr_checksum) 392 return (true); 393 if (res0->hr_compression != res1->hr_compression) 394 return (true); 395 if (res0->hr_timeout != res1->hr_timeout) 396 return (true); 397 if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 398 return (true); 399 /* 400 * When metaflush has changed we don't really need restart, 401 * but it is just easier this way. 402 */ 403 if (res0->hr_metaflush != res1->hr_metaflush) 404 return (true); 405 } 406 return (false); 407 } 408 409 static bool 410 resource_needs_reload(const struct hast_resource *res0, 411 const struct hast_resource *res1) 412 { 413 414 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 415 PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0); 416 PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0); 417 418 if (res0->hr_role != HAST_ROLE_PRIMARY) 419 return (false); 420 421 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 422 return (true); 423 if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0) 424 return (true); 425 if (res0->hr_replication != res1->hr_replication) 426 return (true); 427 if (res0->hr_checksum != res1->hr_checksum) 428 return (true); 429 if (res0->hr_compression != res1->hr_compression) 430 return (true); 431 if (res0->hr_timeout != res1->hr_timeout) 432 return (true); 433 if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 434 return (true); 435 if (res0->hr_metaflush != res1->hr_metaflush) 436 return (true); 437 return (false); 438 } 439 440 static void 441 resource_reload(const struct hast_resource *res) 442 { 443 struct nv *nvin, *nvout; 444 int error; 445 446 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 447 448 nvout = nv_alloc(); 449 nv_add_uint8(nvout, CONTROL_RELOAD, "cmd"); 450 nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr"); 451 nv_add_string(nvout, res->hr_sourceaddr, "sourceaddr"); 452 nv_add_int32(nvout, (int32_t)res->hr_replication, "replication"); 453 nv_add_int32(nvout, (int32_t)res->hr_checksum, "checksum"); 454 nv_add_int32(nvout, (int32_t)res->hr_compression, "compression"); 455 nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout"); 456 nv_add_string(nvout, res->hr_exec, "exec"); 457 nv_add_int32(nvout, (int32_t)res->hr_metaflush, "metaflush"); 458 if (nv_error(nvout) != 0) { 459 nv_free(nvout); 460 pjdlog_error("Unable to allocate header for reload message."); 461 return; 462 } 463 if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) == -1) { 464 pjdlog_errno(LOG_ERR, "Unable to send reload message"); 465 nv_free(nvout); 466 return; 467 } 468 nv_free(nvout); 469 470 /* Receive response. */ 471 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) == -1) { 472 pjdlog_errno(LOG_ERR, "Unable to receive reload reply"); 473 return; 474 } 475 error = nv_get_int16(nvin, "error"); 476 nv_free(nvin); 477 if (error != 0) { 478 pjdlog_common(LOG_ERR, 0, error, "Reload failed"); 479 return; 480 } 481 } 482 483 static void 484 hastd_reload(void) 485 { 486 struct hastd_config *newcfg; 487 struct hast_resource *nres, *cres, *tres; 488 struct hastd_listen *nlst, *clst; 489 struct pidfh *newpfh; 490 unsigned int nlisten; 491 uint8_t role; 492 pid_t otherpid; 493 494 pjdlog_info("Reloading configuration..."); 495 496 newpfh = NULL; 497 498 newcfg = yy_config_parse(cfgpath, false); 499 if (newcfg == NULL) 500 goto failed; 501 502 /* 503 * Check if control address has changed. 504 */ 505 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 506 if (proto_server(newcfg->hc_controladdr, 507 &newcfg->hc_controlconn) == -1) { 508 pjdlog_errno(LOG_ERR, 509 "Unable to listen on control address %s", 510 newcfg->hc_controladdr); 511 goto failed; 512 } 513 } 514 /* 515 * Check if any listen address has changed. 516 */ 517 nlisten = 0; 518 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) { 519 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) { 520 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0) 521 break; 522 } 523 if (clst != NULL && clst->hl_conn != NULL) { 524 pjdlog_info("Keep listening on address %s.", 525 nlst->hl_addr); 526 nlst->hl_conn = clst->hl_conn; 527 nlisten++; 528 } else if (proto_server(nlst->hl_addr, &nlst->hl_conn) == 0) { 529 pjdlog_info("Listening on new address %s.", 530 nlst->hl_addr); 531 nlisten++; 532 } else { 533 pjdlog_errno(LOG_WARNING, 534 "Unable to listen on address %s", nlst->hl_addr); 535 } 536 } 537 if (nlisten == 0) { 538 pjdlog_error("No addresses to listen on."); 539 goto failed; 540 } 541 /* 542 * Check if pidfile's path has changed. 543 */ 544 if (!foreground && pidfile == NULL && 545 strcmp(cfg->hc_pidfile, newcfg->hc_pidfile) != 0) { 546 newpfh = pidfile_open(newcfg->hc_pidfile, 0600, &otherpid); 547 if (newpfh == NULL) { 548 if (errno == EEXIST) { 549 pjdlog_errno(LOG_WARNING, 550 "Another hastd is already running, pidfile: %s, pid: %jd.", 551 newcfg->hc_pidfile, (intmax_t)otherpid); 552 } else { 553 pjdlog_errno(LOG_WARNING, 554 "Unable to open or create pidfile %s", 555 newcfg->hc_pidfile); 556 } 557 } else if (pidfile_write(newpfh) == -1) { 558 /* Write PID to a file. */ 559 pjdlog_errno(LOG_WARNING, 560 "Unable to write PID to file %s", 561 newcfg->hc_pidfile); 562 } else { 563 pjdlog_debug(1, "PID stored in %s.", 564 newcfg->hc_pidfile); 565 } 566 } 567 568 /* No failures from now on. */ 569 570 /* 571 * Switch to new control socket. 572 */ 573 if (newcfg->hc_controlconn != NULL) { 574 pjdlog_info("Control socket changed from %s to %s.", 575 cfg->hc_controladdr, newcfg->hc_controladdr); 576 proto_close(cfg->hc_controlconn); 577 cfg->hc_controlconn = newcfg->hc_controlconn; 578 newcfg->hc_controlconn = NULL; 579 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 580 sizeof(cfg->hc_controladdr)); 581 } 582 /* 583 * Switch to new pidfile. 584 */ 585 if (newpfh != NULL) { 586 pjdlog_info("Pidfile changed from %s to %s.", cfg->hc_pidfile, 587 newcfg->hc_pidfile); 588 (void)pidfile_remove(pfh); 589 pfh = newpfh; 590 (void)strlcpy(cfg->hc_pidfile, newcfg->hc_pidfile, 591 sizeof(cfg->hc_pidfile)); 592 } 593 /* 594 * Switch to new listen addresses. Close all that were removed. 595 */ 596 while ((clst = TAILQ_FIRST(&cfg->hc_listen)) != NULL) { 597 TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) { 598 if (strcmp(nlst->hl_addr, clst->hl_addr) == 0) 599 break; 600 } 601 if (nlst == NULL && clst->hl_conn != NULL) { 602 proto_close(clst->hl_conn); 603 pjdlog_info("No longer listening on address %s.", 604 clst->hl_addr); 605 } 606 TAILQ_REMOVE(&cfg->hc_listen, clst, hl_next); 607 free(clst); 608 } 609 TAILQ_CONCAT(&cfg->hc_listen, &newcfg->hc_listen, hl_next); 610 611 /* 612 * Stop and remove resources that were removed from the configuration. 613 */ 614 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 615 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 616 if (strcmp(cres->hr_name, nres->hr_name) == 0) 617 break; 618 } 619 if (nres == NULL) { 620 control_set_role(cres, HAST_ROLE_INIT); 621 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 622 pjdlog_info("Resource %s removed.", cres->hr_name); 623 free(cres); 624 } 625 } 626 /* 627 * Move new resources to the current configuration. 628 */ 629 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 630 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 631 if (strcmp(cres->hr_name, nres->hr_name) == 0) 632 break; 633 } 634 if (cres == NULL) { 635 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 636 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 637 pjdlog_info("Resource %s added.", nres->hr_name); 638 } 639 } 640 /* 641 * Deal with modified resources. 642 * Depending on what has changed exactly we might want to perform 643 * different actions. 644 * 645 * We do full resource restart in the following situations: 646 * Resource role is INIT or SECONDARY. 647 * Resource role is PRIMARY and path to local component or provider 648 * name has changed. 649 * In case of PRIMARY, the worker process will be killed and restarted, 650 * which also means removing /dev/hast/<name> provider and 651 * recreating it. 652 * 653 * We do just reload (send SIGHUP to worker process) if we act as 654 * PRIMARY, but only if remote address, source address, replication 655 * mode, timeout, execution path or metaflush has changed. 656 * For those, there is no need to restart worker process. 657 * If PRIMARY receives SIGHUP, it will reconnect if remote address or 658 * source address has changed or it will set new timeout if only timeout 659 * has changed or it will update metaflush if only metaflush has 660 * changed. 661 */ 662 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 663 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 664 if (strcmp(cres->hr_name, nres->hr_name) == 0) 665 break; 666 } 667 PJDLOG_ASSERT(cres != NULL); 668 if (resource_needs_restart(cres, nres)) { 669 pjdlog_info("Resource %s configuration was modified, restarting it.", 670 cres->hr_name); 671 role = cres->hr_role; 672 control_set_role(cres, HAST_ROLE_INIT); 673 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 674 free(cres); 675 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 676 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 677 control_set_role(nres, role); 678 } else if (resource_needs_reload(cres, nres)) { 679 pjdlog_info("Resource %s configuration was modified, reloading it.", 680 cres->hr_name); 681 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 682 sizeof(cres->hr_remoteaddr)); 683 strlcpy(cres->hr_sourceaddr, nres->hr_sourceaddr, 684 sizeof(cres->hr_sourceaddr)); 685 cres->hr_replication = nres->hr_replication; 686 cres->hr_checksum = nres->hr_checksum; 687 cres->hr_compression = nres->hr_compression; 688 cres->hr_timeout = nres->hr_timeout; 689 strlcpy(cres->hr_exec, nres->hr_exec, 690 sizeof(cres->hr_exec)); 691 cres->hr_metaflush = nres->hr_metaflush; 692 if (cres->hr_workerpid != 0) 693 resource_reload(cres); 694 } 695 } 696 697 yy_config_free(newcfg); 698 pjdlog_info("Configuration reloaded successfully."); 699 return; 700 failed: 701 if (newcfg != NULL) { 702 if (newcfg->hc_controlconn != NULL) 703 proto_close(newcfg->hc_controlconn); 704 while ((nlst = TAILQ_FIRST(&newcfg->hc_listen)) != NULL) { 705 if (nlst->hl_conn != NULL) { 706 TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) { 707 if (strcmp(nlst->hl_addr, 708 clst->hl_addr) == 0) { 709 break; 710 } 711 } 712 if (clst == NULL || clst->hl_conn == NULL) 713 proto_close(nlst->hl_conn); 714 } 715 TAILQ_REMOVE(&newcfg->hc_listen, nlst, hl_next); 716 free(nlst); 717 } 718 yy_config_free(newcfg); 719 } 720 if (newpfh != NULL) 721 (void)pidfile_remove(newpfh); 722 pjdlog_warning("Configuration not reloaded."); 723 } 724 725 static void 726 terminate_workers(void) 727 { 728 struct hast_resource *res; 729 730 pjdlog_info("Termination signal received, exiting."); 731 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 732 if (res->hr_workerpid == 0) 733 continue; 734 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 735 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 736 if (kill(res->hr_workerpid, SIGTERM) == 0) 737 continue; 738 pjdlog_errno(LOG_WARNING, 739 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 740 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 741 } 742 } 743 744 static void 745 listen_accept(struct hastd_listen *lst) 746 { 747 struct hast_resource *res; 748 struct proto_conn *conn; 749 struct nv *nvin, *nvout, *nverr; 750 const char *resname; 751 const unsigned char *token; 752 char laddr[256], raddr[256]; 753 uint8_t version; 754 size_t size; 755 pid_t pid; 756 int status; 757 758 proto_local_address(lst->hl_conn, laddr, sizeof(laddr)); 759 pjdlog_debug(1, "Accepting connection to %s.", laddr); 760 761 if (proto_accept(lst->hl_conn, &conn) == -1) { 762 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 763 return; 764 } 765 766 proto_local_address(conn, laddr, sizeof(laddr)); 767 proto_remote_address(conn, raddr, sizeof(raddr)); 768 pjdlog_info("Connection from %s to %s.", raddr, laddr); 769 770 /* Error in setting timeout is not critical, but why should it fail? */ 771 if (proto_timeout(conn, HAST_TIMEOUT) == -1) 772 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 773 774 nvin = nvout = nverr = NULL; 775 776 /* 777 * Before receiving any data see if remote host have access to any 778 * resource. 779 */ 780 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 781 if (proto_address_match(conn, res->hr_remoteaddr)) 782 break; 783 } 784 if (res == NULL) { 785 pjdlog_error("Client %s isn't known.", raddr); 786 goto close; 787 } 788 /* Ok, remote host can access at least one resource. */ 789 790 if (hast_proto_recv_hdr(conn, &nvin) == -1) { 791 pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 792 raddr); 793 goto close; 794 } 795 796 resname = nv_get_string(nvin, "resource"); 797 if (resname == NULL) { 798 pjdlog_error("No 'resource' field in the header received from %s.", 799 raddr); 800 goto close; 801 } 802 pjdlog_debug(2, "%s: resource=%s", raddr, resname); 803 version = nv_get_uint8(nvin, "version"); 804 pjdlog_debug(2, "%s: version=%hhu", raddr, version); 805 if (version == 0) { 806 /* 807 * If no version is sent, it means this is protocol version 1. 808 */ 809 version = 1; 810 } 811 token = nv_get_uint8_array(nvin, &size, "token"); 812 /* 813 * NULL token means that this is first connection. 814 */ 815 if (token != NULL && size != sizeof(res->hr_token)) { 816 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 817 raddr, sizeof(res->hr_token), size); 818 goto close; 819 } 820 821 /* 822 * From now on we want to send errors to the remote node. 823 */ 824 nverr = nv_alloc(); 825 826 /* Find resource related to this connection. */ 827 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 828 if (strcmp(resname, res->hr_name) == 0) 829 break; 830 } 831 /* Have we found the resource? */ 832 if (res == NULL) { 833 pjdlog_error("No resource '%s' as requested by %s.", 834 resname, raddr); 835 nv_add_stringf(nverr, "errmsg", "Resource not configured."); 836 goto fail; 837 } 838 839 /* Now that we know resource name setup log prefix. */ 840 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 841 842 /* Does the remote host have access to this resource? */ 843 if (!proto_address_match(conn, res->hr_remoteaddr)) { 844 pjdlog_error("Client %s has no access to the resource.", raddr); 845 nv_add_stringf(nverr, "errmsg", "No access to the resource."); 846 goto fail; 847 } 848 /* Is the resource marked as secondary? */ 849 if (res->hr_role != HAST_ROLE_SECONDARY) { 850 pjdlog_warning("We act as %s for the resource and not as %s as requested by %s.", 851 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 852 raddr); 853 nv_add_stringf(nverr, "errmsg", 854 "Remote node acts as %s for the resource and not as %s.", 855 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 856 if (res->hr_role == HAST_ROLE_PRIMARY) { 857 /* 858 * If we act as primary request the other side to wait 859 * for us a bit, as we might be finishing cleanups. 860 */ 861 nv_add_uint8(nverr, 1, "wait"); 862 } 863 goto fail; 864 } 865 /* Does token (if exists) match? */ 866 if (token != NULL && memcmp(token, res->hr_token, 867 sizeof(res->hr_token)) != 0) { 868 pjdlog_error("Token received from %s doesn't match.", raddr); 869 nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 870 goto fail; 871 } 872 /* 873 * If there is no token, but we have half-open connection 874 * (only remotein) or full connection (worker process is running) 875 * we have to cancel those and accept the new connection. 876 */ 877 if (token == NULL) { 878 PJDLOG_ASSERT(res->hr_remoteout == NULL); 879 pjdlog_debug(1, "Initial connection from %s.", raddr); 880 if (res->hr_workerpid != 0) { 881 PJDLOG_ASSERT(res->hr_remotein == NULL); 882 pjdlog_debug(1, 883 "Worker process exists (pid=%u), stopping it.", 884 (unsigned int)res->hr_workerpid); 885 /* Stop child process. */ 886 if (kill(res->hr_workerpid, SIGINT) == -1) { 887 pjdlog_errno(LOG_ERR, 888 "Unable to stop worker process (pid=%u)", 889 (unsigned int)res->hr_workerpid); 890 /* 891 * Other than logging the problem we 892 * ignore it - nothing smart to do. 893 */ 894 } 895 /* Wait for it to exit. */ 896 else if ((pid = waitpid(res->hr_workerpid, 897 &status, 0)) != res->hr_workerpid) { 898 /* We can only log the problem. */ 899 pjdlog_errno(LOG_ERR, 900 "Waiting for worker process (pid=%u) failed", 901 (unsigned int)res->hr_workerpid); 902 } else { 903 child_exit_log(res->hr_workerpid, status); 904 } 905 child_cleanup(res); 906 } else if (res->hr_remotein != NULL) { 907 char oaddr[256]; 908 909 proto_remote_address(res->hr_remotein, oaddr, 910 sizeof(oaddr)); 911 pjdlog_debug(1, 912 "Canceling half-open connection from %s on connection from %s.", 913 oaddr, raddr); 914 proto_close(res->hr_remotein); 915 res->hr_remotein = NULL; 916 } 917 } 918 919 /* 920 * Checks and cleanups are done. 921 */ 922 923 if (token == NULL) { 924 if (version > HAST_PROTO_VERSION) { 925 pjdlog_info("Remote protocol version %hhu is not supported, falling back to version %hhu.", 926 version, (unsigned char)HAST_PROTO_VERSION); 927 version = HAST_PROTO_VERSION; 928 } 929 pjdlog_debug(1, "Negotiated protocol version %hhu.", version); 930 res->hr_version = version; 931 arc4random_buf(res->hr_token, sizeof(res->hr_token)); 932 nvout = nv_alloc(); 933 nv_add_uint8(nvout, version, "version"); 934 nv_add_uint8_array(nvout, res->hr_token, 935 sizeof(res->hr_token), "token"); 936 if (nv_error(nvout) != 0) { 937 pjdlog_common(LOG_ERR, 0, nv_error(nvout), 938 "Unable to prepare return header for %s", raddr); 939 nv_add_stringf(nverr, "errmsg", 940 "Remote node was unable to prepare return header: %s.", 941 strerror(nv_error(nvout))); 942 goto fail; 943 } 944 if (hast_proto_send(res, conn, nvout, NULL, 0) == -1) { 945 int error = errno; 946 947 pjdlog_errno(LOG_ERR, "Unable to send response to %s", 948 raddr); 949 nv_add_stringf(nverr, "errmsg", 950 "Remote node was unable to send response: %s.", 951 strerror(error)); 952 goto fail; 953 } 954 res->hr_remotein = conn; 955 pjdlog_debug(1, "Incoming connection from %s configured.", 956 raddr); 957 } else { 958 res->hr_remoteout = conn; 959 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 960 hastd_secondary(res, nvin); 961 } 962 nv_free(nvin); 963 nv_free(nvout); 964 nv_free(nverr); 965 pjdlog_prefix_set("%s", ""); 966 return; 967 fail: 968 if (nv_error(nverr) != 0) { 969 pjdlog_common(LOG_ERR, 0, nv_error(nverr), 970 "Unable to prepare error header for %s", raddr); 971 goto close; 972 } 973 if (hast_proto_send(NULL, conn, nverr, NULL, 0) == -1) { 974 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 975 goto close; 976 } 977 close: 978 if (nvin != NULL) 979 nv_free(nvin); 980 if (nvout != NULL) 981 nv_free(nvout); 982 if (nverr != NULL) 983 nv_free(nverr); 984 proto_close(conn); 985 pjdlog_prefix_set("%s", ""); 986 } 987 988 static void 989 connection_migrate(struct hast_resource *res) 990 { 991 struct proto_conn *conn; 992 int16_t val = 0; 993 994 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 995 996 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 997 998 if (proto_recv(res->hr_conn, &val, sizeof(val)) == -1) { 999 pjdlog_errno(LOG_WARNING, 1000 "Unable to receive connection command"); 1001 return; 1002 } 1003 if (proto_client(res->hr_sourceaddr[0] != '\0' ? res->hr_sourceaddr : NULL, 1004 res->hr_remoteaddr, &conn) == -1) { 1005 val = errno; 1006 pjdlog_errno(LOG_WARNING, 1007 "Unable to create outgoing connection to %s", 1008 res->hr_remoteaddr); 1009 goto out; 1010 } 1011 if (proto_connect(conn, -1) == -1) { 1012 val = errno; 1013 pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 1014 res->hr_remoteaddr); 1015 proto_close(conn); 1016 goto out; 1017 } 1018 val = 0; 1019 out: 1020 if (proto_send(res->hr_conn, &val, sizeof(val)) == -1) { 1021 pjdlog_errno(LOG_WARNING, 1022 "Unable to send reply to connection request"); 1023 } 1024 if (val == 0 && proto_connection_send(res->hr_conn, conn) == -1) 1025 pjdlog_errno(LOG_WARNING, "Unable to send connection"); 1026 1027 pjdlog_prefix_set("%s", ""); 1028 } 1029 1030 static void 1031 check_signals(void) 1032 { 1033 struct timespec sigtimeout; 1034 sigset_t mask; 1035 int signo; 1036 1037 sigtimeout.tv_sec = 0; 1038 sigtimeout.tv_nsec = 0; 1039 1040 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1041 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1042 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1043 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1044 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1045 1046 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 1047 switch (signo) { 1048 case SIGINT: 1049 case SIGTERM: 1050 sigexit_received = true; 1051 terminate_workers(); 1052 proto_close(cfg->hc_controlconn); 1053 exit(EX_OK); 1054 break; 1055 case SIGCHLD: 1056 child_exit(); 1057 break; 1058 case SIGHUP: 1059 hastd_reload(); 1060 break; 1061 default: 1062 PJDLOG_ABORT("Unexpected signal (%d).", signo); 1063 } 1064 } 1065 } 1066 1067 static void 1068 main_loop(void) 1069 { 1070 struct hast_resource *res; 1071 struct hastd_listen *lst; 1072 struct timeval seltimeout; 1073 int fd, maxfd, ret; 1074 time_t lastcheck, now; 1075 fd_set rfds; 1076 1077 lastcheck = time(NULL); 1078 seltimeout.tv_sec = REPORT_INTERVAL; 1079 seltimeout.tv_usec = 0; 1080 1081 for (;;) { 1082 check_signals(); 1083 1084 /* Setup descriptors for select(2). */ 1085 FD_ZERO(&rfds); 1086 maxfd = fd = proto_descriptor(cfg->hc_controlconn); 1087 PJDLOG_ASSERT(fd >= 0); 1088 FD_SET(fd, &rfds); 1089 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) { 1090 if (lst->hl_conn == NULL) 1091 continue; 1092 fd = proto_descriptor(lst->hl_conn); 1093 PJDLOG_ASSERT(fd >= 0); 1094 FD_SET(fd, &rfds); 1095 maxfd = MAX(fd, maxfd); 1096 } 1097 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 1098 if (res->hr_event == NULL) 1099 continue; 1100 fd = proto_descriptor(res->hr_event); 1101 PJDLOG_ASSERT(fd >= 0); 1102 FD_SET(fd, &rfds); 1103 maxfd = MAX(fd, maxfd); 1104 if (res->hr_role == HAST_ROLE_PRIMARY) { 1105 /* Only primary workers asks for connections. */ 1106 PJDLOG_ASSERT(res->hr_conn != NULL); 1107 fd = proto_descriptor(res->hr_conn); 1108 PJDLOG_ASSERT(fd >= 0); 1109 FD_SET(fd, &rfds); 1110 maxfd = MAX(fd, maxfd); 1111 } else { 1112 PJDLOG_ASSERT(res->hr_conn == NULL); 1113 } 1114 } 1115 1116 PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE); 1117 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 1118 now = time(NULL); 1119 if (lastcheck + REPORT_INTERVAL <= now) { 1120 hook_check(); 1121 lastcheck = now; 1122 } 1123 if (ret == 0) { 1124 /* 1125 * select(2) timed out, so there should be no 1126 * descriptors to check. 1127 */ 1128 continue; 1129 } else if (ret == -1) { 1130 if (errno == EINTR) 1131 continue; 1132 KEEP_ERRNO((void)pidfile_remove(pfh)); 1133 pjdlog_exit(EX_OSERR, "select() failed"); 1134 } 1135 1136 /* 1137 * Check for signals before we do anything to update our 1138 * info about terminated workers in the meantime. 1139 */ 1140 check_signals(); 1141 1142 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 1143 control_handle(cfg); 1144 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) { 1145 if (lst->hl_conn == NULL) 1146 continue; 1147 if (FD_ISSET(proto_descriptor(lst->hl_conn), &rfds)) 1148 listen_accept(lst); 1149 } 1150 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 1151 if (res->hr_event == NULL) 1152 continue; 1153 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 1154 if (event_recv(res) == 0) 1155 continue; 1156 /* The worker process exited? */ 1157 proto_close(res->hr_event); 1158 res->hr_event = NULL; 1159 if (res->hr_conn != NULL) { 1160 proto_close(res->hr_conn); 1161 res->hr_conn = NULL; 1162 } 1163 continue; 1164 } 1165 if (res->hr_role == HAST_ROLE_PRIMARY) { 1166 PJDLOG_ASSERT(res->hr_conn != NULL); 1167 if (FD_ISSET(proto_descriptor(res->hr_conn), 1168 &rfds)) { 1169 connection_migrate(res); 1170 } 1171 } else { 1172 PJDLOG_ASSERT(res->hr_conn == NULL); 1173 } 1174 } 1175 } 1176 } 1177 1178 static void 1179 dummy_sighandler(int sig __unused) 1180 { 1181 /* Nothing to do. */ 1182 } 1183 1184 int 1185 main(int argc, char *argv[]) 1186 { 1187 struct hastd_listen *lst; 1188 pid_t otherpid; 1189 int debuglevel; 1190 sigset_t mask; 1191 1192 foreground = false; 1193 debuglevel = 0; 1194 1195 for (;;) { 1196 int ch; 1197 1198 ch = getopt(argc, argv, "c:dFhP:"); 1199 if (ch == -1) 1200 break; 1201 switch (ch) { 1202 case 'c': 1203 cfgpath = optarg; 1204 break; 1205 case 'd': 1206 debuglevel++; 1207 break; 1208 case 'F': 1209 foreground = true; 1210 break; 1211 case 'P': 1212 pidfile = optarg; 1213 break; 1214 case 'h': 1215 default: 1216 usage(); 1217 } 1218 } 1219 argc -= optind; 1220 argv += optind; 1221 1222 pjdlog_init(PJDLOG_MODE_STD); 1223 pjdlog_debug_set(debuglevel); 1224 1225 closefrom(MAX(MAX(STDIN_FILENO, STDOUT_FILENO), STDERR_FILENO) + 1); 1226 g_gate_load(); 1227 1228 /* 1229 * When path to the configuration file is relative, obtain full path, 1230 * so we can always find the file, even after daemonizing and changing 1231 * working directory to /. 1232 */ 1233 if (cfgpath[0] != '/') { 1234 const char *newcfgpath; 1235 1236 newcfgpath = realpath(cfgpath, NULL); 1237 if (newcfgpath == NULL) { 1238 pjdlog_exit(EX_CONFIG, 1239 "Unable to obtain full path of %s", cfgpath); 1240 } 1241 cfgpath = newcfgpath; 1242 } 1243 1244 cfg = yy_config_parse(cfgpath, true); 1245 PJDLOG_ASSERT(cfg != NULL); 1246 1247 if (pidfile != NULL) { 1248 if (strlcpy(cfg->hc_pidfile, pidfile, 1249 sizeof(cfg->hc_pidfile)) >= sizeof(cfg->hc_pidfile)) { 1250 pjdlog_exitx(EX_CONFIG, "Pidfile path is too long."); 1251 } 1252 } 1253 1254 if (pidfile != NULL || !foreground) { 1255 pfh = pidfile_open(cfg->hc_pidfile, 0600, &otherpid); 1256 if (pfh == NULL) { 1257 if (errno == EEXIST) { 1258 pjdlog_exitx(EX_TEMPFAIL, 1259 "Another hastd is already running, pidfile: %s, pid: %jd.", 1260 cfg->hc_pidfile, (intmax_t)otherpid); 1261 } 1262 /* 1263 * If we cannot create pidfile for other reasons, 1264 * only warn. 1265 */ 1266 pjdlog_errno(LOG_WARNING, 1267 "Unable to open or create pidfile %s", 1268 cfg->hc_pidfile); 1269 } 1270 } 1271 1272 /* 1273 * Restore default actions for interesting signals in case parent 1274 * process (like init(8)) decided to ignore some of them (like SIGHUP). 1275 */ 1276 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 1277 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 1278 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 1279 /* 1280 * Because SIGCHLD is ignored by default, setup dummy handler for it, 1281 * so we can mask it. 1282 */ 1283 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 1284 1285 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1286 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1287 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1288 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1289 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1290 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 1291 1292 /* Listen on control address. */ 1293 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) == -1) { 1294 KEEP_ERRNO((void)pidfile_remove(pfh)); 1295 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 1296 cfg->hc_controladdr); 1297 } 1298 /* Listen for remote connections. */ 1299 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) { 1300 if (proto_server(lst->hl_addr, &lst->hl_conn) == -1) { 1301 KEEP_ERRNO((void)pidfile_remove(pfh)); 1302 pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 1303 lst->hl_addr); 1304 } 1305 } 1306 1307 if (!foreground) { 1308 if (daemon(0, 0) == -1) { 1309 KEEP_ERRNO((void)pidfile_remove(pfh)); 1310 pjdlog_exit(EX_OSERR, "Unable to daemonize"); 1311 } 1312 1313 /* Start logging to syslog. */ 1314 pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 1315 } 1316 if (pidfile != NULL || !foreground) { 1317 /* Write PID to a file. */ 1318 if (pidfile_write(pfh) == -1) { 1319 pjdlog_errno(LOG_WARNING, 1320 "Unable to write PID to a file %s", 1321 cfg->hc_pidfile); 1322 } else { 1323 pjdlog_debug(1, "PID stored in %s.", cfg->hc_pidfile); 1324 } 1325 } 1326 1327 pjdlog_info("Started successfully, running protocol version %d.", 1328 HAST_PROTO_VERSION); 1329 1330 pjdlog_debug(1, "Listening on control address %s.", 1331 cfg->hc_controladdr); 1332 TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) 1333 pjdlog_info("Listening on address %s.", lst->hl_addr); 1334 1335 hook_init(); 1336 1337 main_loop(); 1338 1339 exit(0); 1340 } 1341