hastd.c revision 218044
1/*- 2 * Copyright (c) 2009-2010 The FreeBSD Foundation 3 * Copyright (c) 2010-2011 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 218044 2011-01-28 21:56:47Z pjd $"); 33 34#include <sys/param.h> 35#include <sys/linker.h> 36#include <sys/module.h> 37#include <sys/stat.h> 38#include <sys/wait.h> 39 40#include <assert.h> 41#include <err.h> 42#include <errno.h> 43#include <libutil.h> 44#include <signal.h> 45#include <stdbool.h> 46#include <stdio.h> 47#include <stdlib.h> 48#include <string.h> 49#include <sysexits.h> 50#include <unistd.h> 51 52#include <activemap.h> 53#include <pjdlog.h> 54 55#include "control.h" 56#include "event.h" 57#include "hast.h" 58#include "hast_proto.h" 59#include "hastd.h" 60#include "hooks.h" 61#include "subr.h" 62 63/* Path to configuration file. */ 64const char *cfgpath = HAST_CONFIG; 65/* Hastd configuration. */ 66static struct hastd_config *cfg; 67/* Was SIGINT or SIGTERM signal received? */ 68bool sigexit_received = false; 69/* PID file handle. */ 70struct pidfh *pfh; 71 72/* How often check for hooks running for too long. */ 73#define REPORT_INTERVAL 5 74 75static void 76usage(void) 77{ 78 79 errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]"); 80} 81 82static void 83g_gate_load(void) 84{ 85 86 if (modfind("g_gate") == -1) { 87 /* Not present in kernel, try loading it. */ 88 if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) { 89 if (errno != EEXIST) { 90 pjdlog_exit(EX_OSERR, 91 "Unable to load geom_gate module"); 92 } 93 } 94 } 95} 96 97void 98descriptors_cleanup(struct hast_resource *res) 99{ 100 struct hast_resource *tres; 101 102 TAILQ_FOREACH(tres, &cfg->hc_resources, hr_next) { 103 if (tres == res) { 104 PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY || 105 (res->hr_remotein == NULL && 106 res->hr_remoteout == NULL)); 107 continue; 108 } 109 if (tres->hr_remotein != NULL) 110 proto_close(tres->hr_remotein); 111 if (tres->hr_remoteout != NULL) 112 proto_close(tres->hr_remoteout); 113 } 114 if (cfg->hc_controlin != NULL) 115 proto_close(cfg->hc_controlin); 116 proto_close(cfg->hc_controlconn); 117 proto_close(cfg->hc_listenconn); 118 (void)pidfile_close(pfh); 119 hook_fini(); 120 pjdlog_fini(); 121} 122 123static const char * 124dtype2str(mode_t mode) 125{ 126 127 if (S_ISBLK(mode)) 128 return ("block device"); 129 else if (S_ISCHR(mode)) 130 return ("character device"); 131 else if (S_ISDIR(mode)) 132 return ("directory"); 133 else if (S_ISFIFO(mode)) 134 return ("pipe or FIFO"); 135 else if (S_ISLNK(mode)) 136 return ("symbolic link"); 137 else if (S_ISREG(mode)) 138 return ("regular file"); 139 else if (S_ISSOCK(mode)) 140 return ("socket"); 141 else if (S_ISWHT(mode)) 142 return ("whiteout"); 143 else 144 return ("unknown"); 145} 146 147void 148descriptors_assert(const struct hast_resource *res, int pjdlogmode) 149{ 150 char msg[256]; 151 struct stat sb; 152 long maxfd; 153 bool isopen; 154 mode_t mode; 155 int fd; 156 157 /* 158 * At this point descriptor to syslog socket is closed, so if we want 159 * to log assertion message, we have to first store it in 'msg' local 160 * buffer and then open syslog socket and log it. 161 */ 162 msg[0] = '\0'; 163 164 maxfd = sysconf(_SC_OPEN_MAX); 165 if (maxfd < 0) { 166 pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed"); 167 maxfd = 16384; 168 } 169 for (fd = 0; fd <= maxfd; fd++) { 170 if (fstat(fd, &sb) == 0) { 171 isopen = true; 172 mode = sb.st_mode; 173 } else if (errno == EBADF) { 174 isopen = false; 175 mode = 0; 176 } else { 177 isopen = true; /* silence gcc */ 178 mode = 0; /* silence gcc */ 179 snprintf(msg, sizeof(msg), 180 "Unable to fstat descriptor %d: %s", fd, 181 strerror(errno)); 182 } 183 if (fd == STDIN_FILENO || fd == STDOUT_FILENO || 184 fd == STDERR_FILENO) { 185 if (!isopen) { 186 snprintf(msg, sizeof(msg), 187 "Descriptor %d (%s) is closed, but should be open.", 188 fd, (fd == STDIN_FILENO ? "stdin" : 189 (fd == STDOUT_FILENO ? "stdout" : "stderr"))); 190 break; 191 } 192 } else if (fd == proto_descriptor(res->hr_event)) { 193 if (!isopen) { 194 snprintf(msg, sizeof(msg), 195 "Descriptor %d (event) is closed, but should be open.", 196 fd); 197 break; 198 } 199 if (!S_ISSOCK(mode)) { 200 snprintf(msg, sizeof(msg), 201 "Descriptor %d (event) is %s, but should be %s.", 202 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 203 break; 204 } 205 } else if (fd == proto_descriptor(res->hr_ctrl)) { 206 if (!isopen) { 207 snprintf(msg, sizeof(msg), 208 "Descriptor %d (ctrl) is closed, but should be open.", 209 fd); 210 break; 211 } 212 if (!S_ISSOCK(mode)) { 213 snprintf(msg, sizeof(msg), 214 "Descriptor %d (ctrl) is %s, but should be %s.", 215 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 216 break; 217 } 218 } else if (res->hr_role == HAST_ROLE_SECONDARY && 219 fd == proto_descriptor(res->hr_remotein)) { 220 if (!isopen) { 221 snprintf(msg, sizeof(msg), 222 "Descriptor %d (remote in) is closed, but should be open.", 223 fd); 224 break; 225 } 226 if (!S_ISSOCK(mode)) { 227 snprintf(msg, sizeof(msg), 228 "Descriptor %d (remote in) is %s, but should be %s.", 229 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 230 break; 231 } 232 } else if (res->hr_role == HAST_ROLE_SECONDARY && 233 fd == proto_descriptor(res->hr_remoteout)) { 234 if (!isopen) { 235 snprintf(msg, sizeof(msg), 236 "Descriptor %d (remote out) is closed, but should be open.", 237 fd); 238 break; 239 } 240 if (!S_ISSOCK(mode)) { 241 snprintf(msg, sizeof(msg), 242 "Descriptor %d (remote out) is %s, but should be %s.", 243 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 244 break; 245 } 246 } else { 247 if (isopen) { 248 snprintf(msg, sizeof(msg), 249 "Descriptor %d is open (%s), but should be closed.", 250 fd, dtype2str(mode)); 251 break; 252 } 253 } 254 } 255 if (msg[0] != '\0') { 256 pjdlog_init(pjdlogmode); 257 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 258 role2str(res->hr_role)); 259 PJDLOG_ABORT("%s", msg); 260 } 261} 262 263static void 264child_exit_log(unsigned int pid, int status) 265{ 266 267 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 268 pjdlog_debug(1, "Worker process exited gracefully (pid=%u).", 269 pid); 270 } else if (WIFSIGNALED(status)) { 271 pjdlog_error("Worker process killed (pid=%u, signal=%d).", 272 pid, WTERMSIG(status)); 273 } else { 274 pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).", 275 pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1); 276 } 277} 278 279static void 280child_exit(void) 281{ 282 struct hast_resource *res; 283 int status; 284 pid_t pid; 285 286 while ((pid = wait3(&status, WNOHANG, NULL)) > 0) { 287 /* Find resource related to the process that just exited. */ 288 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 289 if (pid == res->hr_workerpid) 290 break; 291 } 292 if (res == NULL) { 293 /* 294 * This can happen when new connection arrives and we 295 * cancel child responsible for the old one or if this 296 * was hook which we executed. 297 */ 298 hook_check_one(pid, status); 299 continue; 300 } 301 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 302 role2str(res->hr_role)); 303 child_exit_log(pid, status); 304 child_cleanup(res); 305 if (res->hr_role == HAST_ROLE_PRIMARY) { 306 /* 307 * Restart child process if it was killed by signal 308 * or exited because of temporary problem. 309 */ 310 if (WIFSIGNALED(status) || 311 (WIFEXITED(status) && 312 WEXITSTATUS(status) == EX_TEMPFAIL)) { 313 sleep(1); 314 pjdlog_info("Restarting worker process."); 315 hastd_primary(res); 316 } else { 317 res->hr_role = HAST_ROLE_INIT; 318 pjdlog_info("Changing resource role back to %s.", 319 role2str(res->hr_role)); 320 } 321 } 322 pjdlog_prefix_set("%s", ""); 323 } 324} 325 326static bool 327resource_needs_restart(const struct hast_resource *res0, 328 const struct hast_resource *res1) 329{ 330 331 assert(strcmp(res0->hr_name, res1->hr_name) == 0); 332 333 if (strcmp(res0->hr_provname, res1->hr_provname) != 0) 334 return (true); 335 if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0) 336 return (true); 337 if (res0->hr_role == HAST_ROLE_INIT || 338 res0->hr_role == HAST_ROLE_SECONDARY) { 339 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 340 return (true); 341 if (res0->hr_replication != res1->hr_replication) 342 return (true); 343 if (res0->hr_timeout != res1->hr_timeout) 344 return (true); 345 if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 346 return (true); 347 } 348 return (false); 349} 350 351static bool 352resource_needs_reload(const struct hast_resource *res0, 353 const struct hast_resource *res1) 354{ 355 356 assert(strcmp(res0->hr_name, res1->hr_name) == 0); 357 assert(strcmp(res0->hr_provname, res1->hr_provname) == 0); 358 assert(strcmp(res0->hr_localpath, res1->hr_localpath) == 0); 359 360 if (res0->hr_role != HAST_ROLE_PRIMARY) 361 return (false); 362 363 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 364 return (true); 365 if (res0->hr_replication != res1->hr_replication) 366 return (true); 367 if (res0->hr_timeout != res1->hr_timeout) 368 return (true); 369 if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 370 return (true); 371 return (false); 372} 373 374static void 375resource_reload(const struct hast_resource *res) 376{ 377 struct nv *nvin, *nvout; 378 int error; 379 380 assert(res->hr_role == HAST_ROLE_PRIMARY); 381 382 nvout = nv_alloc(); 383 nv_add_uint8(nvout, HASTCTL_RELOAD, "cmd"); 384 nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr"); 385 nv_add_int32(nvout, (int32_t)res->hr_replication, "replication"); 386 nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout"); 387 nv_add_string(nvout, res->hr_exec, "exec"); 388 if (nv_error(nvout) != 0) { 389 nv_free(nvout); 390 pjdlog_error("Unable to allocate header for reload message."); 391 return; 392 } 393 if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) { 394 pjdlog_errno(LOG_ERR, "Unable to send reload message"); 395 nv_free(nvout); 396 return; 397 } 398 nv_free(nvout); 399 400 /* Receive response. */ 401 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) { 402 pjdlog_errno(LOG_ERR, "Unable to receive reload reply"); 403 return; 404 } 405 error = nv_get_int16(nvin, "error"); 406 nv_free(nvin); 407 if (error != 0) { 408 pjdlog_common(LOG_ERR, 0, error, "Reload failed"); 409 return; 410 } 411} 412 413static void 414hastd_reload(void) 415{ 416 struct hastd_config *newcfg; 417 struct hast_resource *nres, *cres, *tres; 418 uint8_t role; 419 420 pjdlog_info("Reloading configuration..."); 421 422 newcfg = yy_config_parse(cfgpath, false); 423 if (newcfg == NULL) 424 goto failed; 425 426 /* 427 * Check if control address has changed. 428 */ 429 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 430 if (proto_server(newcfg->hc_controladdr, 431 &newcfg->hc_controlconn) < 0) { 432 pjdlog_errno(LOG_ERR, 433 "Unable to listen on control address %s", 434 newcfg->hc_controladdr); 435 goto failed; 436 } 437 } 438 /* 439 * Check if listen address has changed. 440 */ 441 if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) { 442 if (proto_server(newcfg->hc_listenaddr, 443 &newcfg->hc_listenconn) < 0) { 444 pjdlog_errno(LOG_ERR, "Unable to listen on address %s", 445 newcfg->hc_listenaddr); 446 goto failed; 447 } 448 } 449 /* 450 * Only when both control and listen sockets are successfully 451 * initialized switch them to new configuration. 452 */ 453 if (newcfg->hc_controlconn != NULL) { 454 pjdlog_info("Control socket changed from %s to %s.", 455 cfg->hc_controladdr, newcfg->hc_controladdr); 456 proto_close(cfg->hc_controlconn); 457 cfg->hc_controlconn = newcfg->hc_controlconn; 458 newcfg->hc_controlconn = NULL; 459 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 460 sizeof(cfg->hc_controladdr)); 461 } 462 if (newcfg->hc_listenconn != NULL) { 463 pjdlog_info("Listen socket changed from %s to %s.", 464 cfg->hc_listenaddr, newcfg->hc_listenaddr); 465 proto_close(cfg->hc_listenconn); 466 cfg->hc_listenconn = newcfg->hc_listenconn; 467 newcfg->hc_listenconn = NULL; 468 strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr, 469 sizeof(cfg->hc_listenaddr)); 470 } 471 472 /* 473 * Stop and remove resources that were removed from the configuration. 474 */ 475 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 476 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 477 if (strcmp(cres->hr_name, nres->hr_name) == 0) 478 break; 479 } 480 if (nres == NULL) { 481 control_set_role(cres, HAST_ROLE_INIT); 482 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 483 pjdlog_info("Resource %s removed.", cres->hr_name); 484 free(cres); 485 } 486 } 487 /* 488 * Move new resources to the current configuration. 489 */ 490 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 491 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 492 if (strcmp(cres->hr_name, nres->hr_name) == 0) 493 break; 494 } 495 if (cres == NULL) { 496 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 497 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 498 pjdlog_info("Resource %s added.", nres->hr_name); 499 } 500 } 501 /* 502 * Deal with modified resources. 503 * Depending on what has changed exactly we might want to perform 504 * different actions. 505 * 506 * We do full resource restart in the following situations: 507 * Resource role is INIT or SECONDARY. 508 * Resource role is PRIMARY and path to local component or provider 509 * name has changed. 510 * In case of PRIMARY, the worker process will be killed and restarted, 511 * which also means removing /dev/hast/<name> provider and 512 * recreating it. 513 * 514 * We do just reload (send SIGHUP to worker process) if we act as 515 * PRIMARY, but only if remote address, replication mode, timeout or 516 * execution path has changed. For those, there is no need to restart 517 * worker process. 518 * If PRIMARY receives SIGHUP, it will reconnect if remote address or 519 * replication mode has changed or simply set new timeout if only 520 * timeout has changed. 521 */ 522 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 523 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 524 if (strcmp(cres->hr_name, nres->hr_name) == 0) 525 break; 526 } 527 assert(cres != NULL); 528 if (resource_needs_restart(cres, nres)) { 529 pjdlog_info("Resource %s configuration was modified, restarting it.", 530 cres->hr_name); 531 role = cres->hr_role; 532 control_set_role(cres, HAST_ROLE_INIT); 533 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 534 free(cres); 535 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 536 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 537 control_set_role(nres, role); 538 } else if (resource_needs_reload(cres, nres)) { 539 pjdlog_info("Resource %s configuration was modified, reloading it.", 540 cres->hr_name); 541 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 542 sizeof(cres->hr_remoteaddr)); 543 cres->hr_replication = nres->hr_replication; 544 cres->hr_timeout = nres->hr_timeout; 545 strlcpy(cres->hr_exec, nres->hr_exec, 546 sizeof(cres->hr_exec)); 547 if (cres->hr_workerpid != 0) 548 resource_reload(cres); 549 } 550 } 551 552 yy_config_free(newcfg); 553 pjdlog_info("Configuration reloaded successfully."); 554 return; 555failed: 556 if (newcfg != NULL) { 557 if (newcfg->hc_controlconn != NULL) 558 proto_close(newcfg->hc_controlconn); 559 if (newcfg->hc_listenconn != NULL) 560 proto_close(newcfg->hc_listenconn); 561 yy_config_free(newcfg); 562 } 563 pjdlog_warning("Configuration not reloaded."); 564} 565 566static void 567terminate_workers(void) 568{ 569 struct hast_resource *res; 570 571 pjdlog_info("Termination signal received, exiting."); 572 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 573 if (res->hr_workerpid == 0) 574 continue; 575 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 576 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 577 if (kill(res->hr_workerpid, SIGTERM) == 0) 578 continue; 579 pjdlog_errno(LOG_WARNING, 580 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 581 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 582 } 583} 584 585static void 586listen_accept(void) 587{ 588 struct hast_resource *res; 589 struct proto_conn *conn; 590 struct nv *nvin, *nvout, *nverr; 591 const char *resname; 592 const unsigned char *token; 593 char laddr[256], raddr[256]; 594 size_t size; 595 pid_t pid; 596 int status; 597 598 proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr)); 599 pjdlog_debug(1, "Accepting connection to %s.", laddr); 600 601 if (proto_accept(cfg->hc_listenconn, &conn) < 0) { 602 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 603 return; 604 } 605 606 proto_local_address(conn, laddr, sizeof(laddr)); 607 proto_remote_address(conn, raddr, sizeof(raddr)); 608 pjdlog_info("Connection from %s to %s.", raddr, laddr); 609 610 /* Error in setting timeout is not critical, but why should it fail? */ 611 if (proto_timeout(conn, HAST_TIMEOUT) < 0) 612 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 613 614 nvin = nvout = nverr = NULL; 615 616 /* 617 * Before receiving any data see if remote host have access to any 618 * resource. 619 */ 620 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 621 if (proto_address_match(conn, res->hr_remoteaddr)) 622 break; 623 } 624 if (res == NULL) { 625 pjdlog_error("Client %s isn't known.", raddr); 626 goto close; 627 } 628 /* Ok, remote host can access at least one resource. */ 629 630 if (hast_proto_recv_hdr(conn, &nvin) < 0) { 631 pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 632 raddr); 633 goto close; 634 } 635 636 resname = nv_get_string(nvin, "resource"); 637 if (resname == NULL) { 638 pjdlog_error("No 'resource' field in the header received from %s.", 639 raddr); 640 goto close; 641 } 642 pjdlog_debug(2, "%s: resource=%s", raddr, resname); 643 token = nv_get_uint8_array(nvin, &size, "token"); 644 /* 645 * NULL token means that this is first conection. 646 */ 647 if (token != NULL && size != sizeof(res->hr_token)) { 648 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 649 raddr, sizeof(res->hr_token), size); 650 goto close; 651 } 652 653 /* 654 * From now on we want to send errors to the remote node. 655 */ 656 nverr = nv_alloc(); 657 658 /* Find resource related to this connection. */ 659 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 660 if (strcmp(resname, res->hr_name) == 0) 661 break; 662 } 663 /* Have we found the resource? */ 664 if (res == NULL) { 665 pjdlog_error("No resource '%s' as requested by %s.", 666 resname, raddr); 667 nv_add_stringf(nverr, "errmsg", "Resource not configured."); 668 goto fail; 669 } 670 671 /* Now that we know resource name setup log prefix. */ 672 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 673 674 /* Does the remote host have access to this resource? */ 675 if (!proto_address_match(conn, res->hr_remoteaddr)) { 676 pjdlog_error("Client %s has no access to the resource.", raddr); 677 nv_add_stringf(nverr, "errmsg", "No access to the resource."); 678 goto fail; 679 } 680 /* Is the resource marked as secondary? */ 681 if (res->hr_role != HAST_ROLE_SECONDARY) { 682 pjdlog_error("We act as %s for the resource and not as %s as requested by %s.", 683 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 684 raddr); 685 nv_add_stringf(nverr, "errmsg", 686 "Remote node acts as %s for the resource and not as %s.", 687 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 688 goto fail; 689 } 690 /* Does token (if exists) match? */ 691 if (token != NULL && memcmp(token, res->hr_token, 692 sizeof(res->hr_token)) != 0) { 693 pjdlog_error("Token received from %s doesn't match.", raddr); 694 nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 695 goto fail; 696 } 697 /* 698 * If there is no token, but we have half-open connection 699 * (only remotein) or full connection (worker process is running) 700 * we have to cancel those and accept the new connection. 701 */ 702 if (token == NULL) { 703 assert(res->hr_remoteout == NULL); 704 pjdlog_debug(1, "Initial connection from %s.", raddr); 705 if (res->hr_workerpid != 0) { 706 assert(res->hr_remotein == NULL); 707 pjdlog_debug(1, 708 "Worker process exists (pid=%u), stopping it.", 709 (unsigned int)res->hr_workerpid); 710 /* Stop child process. */ 711 if (kill(res->hr_workerpid, SIGINT) < 0) { 712 pjdlog_errno(LOG_ERR, 713 "Unable to stop worker process (pid=%u)", 714 (unsigned int)res->hr_workerpid); 715 /* 716 * Other than logging the problem we 717 * ignore it - nothing smart to do. 718 */ 719 } 720 /* Wait for it to exit. */ 721 else if ((pid = waitpid(res->hr_workerpid, 722 &status, 0)) != res->hr_workerpid) { 723 /* We can only log the problem. */ 724 pjdlog_errno(LOG_ERR, 725 "Waiting for worker process (pid=%u) failed", 726 (unsigned int)res->hr_workerpid); 727 } else { 728 child_exit_log(res->hr_workerpid, status); 729 } 730 child_cleanup(res); 731 } else if (res->hr_remotein != NULL) { 732 char oaddr[256]; 733 734 proto_remote_address(res->hr_remotein, oaddr, 735 sizeof(oaddr)); 736 pjdlog_debug(1, 737 "Canceling half-open connection from %s on connection from %s.", 738 oaddr, raddr); 739 proto_close(res->hr_remotein); 740 res->hr_remotein = NULL; 741 } 742 } 743 744 /* 745 * Checks and cleanups are done. 746 */ 747 748 if (token == NULL) { 749 arc4random_buf(res->hr_token, sizeof(res->hr_token)); 750 nvout = nv_alloc(); 751 nv_add_uint8_array(nvout, res->hr_token, 752 sizeof(res->hr_token), "token"); 753 if (nv_error(nvout) != 0) { 754 pjdlog_common(LOG_ERR, 0, nv_error(nvout), 755 "Unable to prepare return header for %s", raddr); 756 nv_add_stringf(nverr, "errmsg", 757 "Remote node was unable to prepare return header: %s.", 758 strerror(nv_error(nvout))); 759 goto fail; 760 } 761 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) { 762 int error = errno; 763 764 pjdlog_errno(LOG_ERR, "Unable to send response to %s", 765 raddr); 766 nv_add_stringf(nverr, "errmsg", 767 "Remote node was unable to send response: %s.", 768 strerror(error)); 769 goto fail; 770 } 771 res->hr_remotein = conn; 772 pjdlog_debug(1, "Incoming connection from %s configured.", 773 raddr); 774 } else { 775 res->hr_remoteout = conn; 776 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 777 hastd_secondary(res, nvin); 778 } 779 nv_free(nvin); 780 nv_free(nvout); 781 nv_free(nverr); 782 pjdlog_prefix_set("%s", ""); 783 return; 784fail: 785 if (nv_error(nverr) != 0) { 786 pjdlog_common(LOG_ERR, 0, nv_error(nverr), 787 "Unable to prepare error header for %s", raddr); 788 goto close; 789 } 790 if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) { 791 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 792 goto close; 793 } 794close: 795 if (nvin != NULL) 796 nv_free(nvin); 797 if (nvout != NULL) 798 nv_free(nvout); 799 if (nverr != NULL) 800 nv_free(nverr); 801 proto_close(conn); 802 pjdlog_prefix_set("%s", ""); 803} 804 805static void 806main_loop(void) 807{ 808 struct hast_resource *res; 809 struct timeval seltimeout; 810 struct timespec sigtimeout; 811 int fd, maxfd, ret, signo; 812 sigset_t mask; 813 fd_set rfds; 814 815 seltimeout.tv_sec = REPORT_INTERVAL; 816 seltimeout.tv_usec = 0; 817 sigtimeout.tv_sec = 0; 818 sigtimeout.tv_nsec = 0; 819 820 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 821 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 822 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 823 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 824 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 825 826 pjdlog_info("Started successfully, running protocol version %d.", 827 HAST_PROTO_VERSION); 828 829 for (;;) { 830 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 831 switch (signo) { 832 case SIGINT: 833 case SIGTERM: 834 sigexit_received = true; 835 terminate_workers(); 836 proto_close(cfg->hc_controlconn); 837 exit(EX_OK); 838 break; 839 case SIGCHLD: 840 child_exit(); 841 break; 842 case SIGHUP: 843 hastd_reload(); 844 break; 845 default: 846 assert(!"invalid condition"); 847 } 848 } 849 850 /* Setup descriptors for select(2). */ 851 FD_ZERO(&rfds); 852 maxfd = fd = proto_descriptor(cfg->hc_controlconn); 853 assert(fd >= 0); 854 FD_SET(fd, &rfds); 855 fd = proto_descriptor(cfg->hc_listenconn); 856 assert(fd >= 0); 857 FD_SET(fd, &rfds); 858 maxfd = fd > maxfd ? fd : maxfd; 859 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 860 if (res->hr_event == NULL) 861 continue; 862 fd = proto_descriptor(res->hr_event); 863 assert(fd >= 0); 864 FD_SET(fd, &rfds); 865 maxfd = fd > maxfd ? fd : maxfd; 866 } 867 868 assert(maxfd + 1 <= (int)FD_SETSIZE); 869 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 870 if (ret == 0) 871 hook_check(); 872 else if (ret == -1) { 873 if (errno == EINTR) 874 continue; 875 KEEP_ERRNO((void)pidfile_remove(pfh)); 876 pjdlog_exit(EX_OSERR, "select() failed"); 877 } 878 879 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 880 control_handle(cfg); 881 if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds)) 882 listen_accept(); 883 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 884 if (res->hr_event == NULL) 885 continue; 886 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 887 if (event_recv(res) == 0) 888 continue; 889 /* The worker process exited? */ 890 proto_close(res->hr_event); 891 res->hr_event = NULL; 892 } 893 } 894 } 895} 896 897static void 898dummy_sighandler(int sig __unused) 899{ 900 /* Nothing to do. */ 901} 902 903int 904main(int argc, char *argv[]) 905{ 906 const char *pidfile; 907 pid_t otherpid; 908 bool foreground; 909 int debuglevel; 910 sigset_t mask; 911 912 foreground = false; 913 debuglevel = 0; 914 pidfile = HASTD_PIDFILE; 915 916 for (;;) { 917 int ch; 918 919 ch = getopt(argc, argv, "c:dFhP:"); 920 if (ch == -1) 921 break; 922 switch (ch) { 923 case 'c': 924 cfgpath = optarg; 925 break; 926 case 'd': 927 debuglevel++; 928 break; 929 case 'F': 930 foreground = true; 931 break; 932 case 'P': 933 pidfile = optarg; 934 break; 935 case 'h': 936 default: 937 usage(); 938 } 939 } 940 argc -= optind; 941 argv += optind; 942 943 pjdlog_init(PJDLOG_MODE_STD); 944 pjdlog_debug_set(debuglevel); 945 946 g_gate_load(); 947 948 pfh = pidfile_open(pidfile, 0600, &otherpid); 949 if (pfh == NULL) { 950 if (errno == EEXIST) { 951 pjdlog_exitx(EX_TEMPFAIL, 952 "Another hastd is already running, pid: %jd.", 953 (intmax_t)otherpid); 954 } 955 /* If we cannot create pidfile from other reasons, only warn. */ 956 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile"); 957 } 958 959 cfg = yy_config_parse(cfgpath, true); 960 assert(cfg != NULL); 961 962 /* 963 * Restore default actions for interesting signals in case parent 964 * process (like init(8)) decided to ignore some of them (like SIGHUP). 965 */ 966 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 967 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 968 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 969 /* 970 * Because SIGCHLD is ignored by default, setup dummy handler for it, 971 * so we can mask it. 972 */ 973 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 974 975 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 976 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 977 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 978 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 979 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 980 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 981 982 /* Listen on control address. */ 983 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) { 984 KEEP_ERRNO((void)pidfile_remove(pfh)); 985 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 986 cfg->hc_controladdr); 987 } 988 /* Listen for remote connections. */ 989 if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) { 990 KEEP_ERRNO((void)pidfile_remove(pfh)); 991 pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 992 cfg->hc_listenaddr); 993 } 994 995 if (!foreground) { 996 if (daemon(0, 0) < 0) { 997 KEEP_ERRNO((void)pidfile_remove(pfh)); 998 pjdlog_exit(EX_OSERR, "Unable to daemonize"); 999 } 1000 1001 /* Start logging to syslog. */ 1002 pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 1003 1004 /* Write PID to a file. */ 1005 if (pidfile_write(pfh) < 0) { 1006 pjdlog_errno(LOG_WARNING, 1007 "Unable to write PID to a file"); 1008 } 1009 } 1010 1011 hook_init(); 1012 1013 main_loop(); 1014 1015 exit(0); 1016} 1017