hastd.c revision 218218
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 218218 2011-02-03 11:39:49Z 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 <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 96void 97descriptors_cleanup(struct hast_resource *res) 98{ 99 struct hast_resource *tres; 100 101 TAILQ_FOREACH(tres, &cfg->hc_resources, hr_next) { 102 if (tres == res) { 103 PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY || 104 (res->hr_remotein == NULL && 105 res->hr_remoteout == NULL)); 106 continue; 107 } 108 if (tres->hr_remotein != NULL) 109 proto_close(tres->hr_remotein); 110 if (tres->hr_remoteout != NULL) 111 proto_close(tres->hr_remoteout); 112 } 113 if (cfg->hc_controlin != NULL) 114 proto_close(cfg->hc_controlin); 115 proto_close(cfg->hc_controlconn); 116 proto_close(cfg->hc_listenconn); 117 (void)pidfile_close(pfh); 118 hook_fini(); 119 pjdlog_fini(); 120} 121 122static const char * 123dtype2str(mode_t mode) 124{ 125 126 if (S_ISBLK(mode)) 127 return ("block device"); 128 else if (S_ISCHR(mode)) 129 return ("character device"); 130 else if (S_ISDIR(mode)) 131 return ("directory"); 132 else if (S_ISFIFO(mode)) 133 return ("pipe or FIFO"); 134 else if (S_ISLNK(mode)) 135 return ("symbolic link"); 136 else if (S_ISREG(mode)) 137 return ("regular file"); 138 else if (S_ISSOCK(mode)) 139 return ("socket"); 140 else if (S_ISWHT(mode)) 141 return ("whiteout"); 142 else 143 return ("unknown"); 144} 145 146void 147descriptors_assert(const struct hast_resource *res, int pjdlogmode) 148{ 149 char msg[256]; 150 struct stat sb; 151 long maxfd; 152 bool isopen; 153 mode_t mode; 154 int fd; 155 156 /* 157 * At this point descriptor to syslog socket is closed, so if we want 158 * to log assertion message, we have to first store it in 'msg' local 159 * buffer and then open syslog socket and log it. 160 */ 161 msg[0] = '\0'; 162 163 maxfd = sysconf(_SC_OPEN_MAX); 164 if (maxfd < 0) { 165 pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed"); 166 maxfd = 16384; 167 } 168 for (fd = 0; fd <= maxfd; fd++) { 169 if (fstat(fd, &sb) == 0) { 170 isopen = true; 171 mode = sb.st_mode; 172 } else if (errno == EBADF) { 173 isopen = false; 174 mode = 0; 175 } else { 176 isopen = true; /* silence gcc */ 177 mode = 0; /* silence gcc */ 178 snprintf(msg, sizeof(msg), 179 "Unable to fstat descriptor %d: %s", fd, 180 strerror(errno)); 181 } 182 if (fd == STDIN_FILENO || fd == STDOUT_FILENO || 183 fd == STDERR_FILENO) { 184 if (!isopen) { 185 snprintf(msg, sizeof(msg), 186 "Descriptor %d (%s) is closed, but should be open.", 187 fd, (fd == STDIN_FILENO ? "stdin" : 188 (fd == STDOUT_FILENO ? "stdout" : "stderr"))); 189 break; 190 } 191 } else if (fd == proto_descriptor(res->hr_event)) { 192 if (!isopen) { 193 snprintf(msg, sizeof(msg), 194 "Descriptor %d (event) is closed, but should be open.", 195 fd); 196 break; 197 } 198 if (!S_ISSOCK(mode)) { 199 snprintf(msg, sizeof(msg), 200 "Descriptor %d (event) is %s, but should be %s.", 201 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 202 break; 203 } 204 } else if (fd == proto_descriptor(res->hr_ctrl)) { 205 if (!isopen) { 206 snprintf(msg, sizeof(msg), 207 "Descriptor %d (ctrl) is closed, but should be open.", 208 fd); 209 break; 210 } 211 if (!S_ISSOCK(mode)) { 212 snprintf(msg, sizeof(msg), 213 "Descriptor %d (ctrl) is %s, but should be %s.", 214 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 215 break; 216 } 217 } else if (fd == proto_descriptor(res->hr_conn)) { 218 if (!isopen) { 219 snprintf(msg, sizeof(msg), 220 "Descriptor %d (conn) is closed, but should be open.", 221 fd); 222 break; 223 } 224 if (!S_ISSOCK(mode)) { 225 snprintf(msg, sizeof(msg), 226 "Descriptor %d (conn) is %s, but should be %s.", 227 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 228 break; 229 } 230 } else if (res->hr_role == HAST_ROLE_SECONDARY && 231 fd == proto_descriptor(res->hr_remotein)) { 232 if (!isopen) { 233 snprintf(msg, sizeof(msg), 234 "Descriptor %d (remote in) is closed, but should be open.", 235 fd); 236 break; 237 } 238 if (!S_ISSOCK(mode)) { 239 snprintf(msg, sizeof(msg), 240 "Descriptor %d (remote in) is %s, but should be %s.", 241 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 242 break; 243 } 244 } else if (res->hr_role == HAST_ROLE_SECONDARY && 245 fd == proto_descriptor(res->hr_remoteout)) { 246 if (!isopen) { 247 snprintf(msg, sizeof(msg), 248 "Descriptor %d (remote out) is closed, but should be open.", 249 fd); 250 break; 251 } 252 if (!S_ISSOCK(mode)) { 253 snprintf(msg, sizeof(msg), 254 "Descriptor %d (remote out) is %s, but should be %s.", 255 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 256 break; 257 } 258 } else { 259 if (isopen) { 260 snprintf(msg, sizeof(msg), 261 "Descriptor %d is open (%s), but should be closed.", 262 fd, dtype2str(mode)); 263 break; 264 } 265 } 266 } 267 if (msg[0] != '\0') { 268 pjdlog_init(pjdlogmode); 269 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 270 role2str(res->hr_role)); 271 PJDLOG_ABORT("%s", msg); 272 } 273} 274 275static void 276child_exit_log(unsigned int pid, int status) 277{ 278 279 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 280 pjdlog_debug(1, "Worker process exited gracefully (pid=%u).", 281 pid); 282 } else if (WIFSIGNALED(status)) { 283 pjdlog_error("Worker process killed (pid=%u, signal=%d).", 284 pid, WTERMSIG(status)); 285 } else { 286 pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).", 287 pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1); 288 } 289} 290 291static void 292child_exit(void) 293{ 294 struct hast_resource *res; 295 int status; 296 pid_t pid; 297 298 while ((pid = wait3(&status, WNOHANG, NULL)) > 0) { 299 /* Find resource related to the process that just exited. */ 300 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 301 if (pid == res->hr_workerpid) 302 break; 303 } 304 if (res == NULL) { 305 /* 306 * This can happen when new connection arrives and we 307 * cancel child responsible for the old one or if this 308 * was hook which we executed. 309 */ 310 hook_check_one(pid, status); 311 continue; 312 } 313 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 314 role2str(res->hr_role)); 315 child_exit_log(pid, status); 316 child_cleanup(res); 317 if (res->hr_role == HAST_ROLE_PRIMARY) { 318 /* 319 * Restart child process if it was killed by signal 320 * or exited because of temporary problem. 321 */ 322 if (WIFSIGNALED(status) || 323 (WIFEXITED(status) && 324 WEXITSTATUS(status) == EX_TEMPFAIL)) { 325 sleep(1); 326 pjdlog_info("Restarting worker process."); 327 hastd_primary(res); 328 } else { 329 res->hr_role = HAST_ROLE_INIT; 330 pjdlog_info("Changing resource role back to %s.", 331 role2str(res->hr_role)); 332 } 333 } 334 pjdlog_prefix_set("%s", ""); 335 } 336} 337 338static bool 339resource_needs_restart(const struct hast_resource *res0, 340 const struct hast_resource *res1) 341{ 342 343 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 344 345 if (strcmp(res0->hr_provname, res1->hr_provname) != 0) 346 return (true); 347 if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0) 348 return (true); 349 if (res0->hr_role == HAST_ROLE_INIT || 350 res0->hr_role == HAST_ROLE_SECONDARY) { 351 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 352 return (true); 353 if (res0->hr_replication != res1->hr_replication) 354 return (true); 355 if (res0->hr_timeout != res1->hr_timeout) 356 return (true); 357 if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 358 return (true); 359 } 360 return (false); 361} 362 363static bool 364resource_needs_reload(const struct hast_resource *res0, 365 const struct hast_resource *res1) 366{ 367 368 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 369 PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0); 370 PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0); 371 372 if (res0->hr_role != HAST_ROLE_PRIMARY) 373 return (false); 374 375 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 376 return (true); 377 if (res0->hr_replication != res1->hr_replication) 378 return (true); 379 if (res0->hr_timeout != res1->hr_timeout) 380 return (true); 381 if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 382 return (true); 383 return (false); 384} 385 386static void 387resource_reload(const struct hast_resource *res) 388{ 389 struct nv *nvin, *nvout; 390 int error; 391 392 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 393 394 nvout = nv_alloc(); 395 nv_add_uint8(nvout, HASTCTL_RELOAD, "cmd"); 396 nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr"); 397 nv_add_int32(nvout, (int32_t)res->hr_replication, "replication"); 398 nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout"); 399 nv_add_string(nvout, res->hr_exec, "exec"); 400 if (nv_error(nvout) != 0) { 401 nv_free(nvout); 402 pjdlog_error("Unable to allocate header for reload message."); 403 return; 404 } 405 if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) { 406 pjdlog_errno(LOG_ERR, "Unable to send reload message"); 407 nv_free(nvout); 408 return; 409 } 410 nv_free(nvout); 411 412 /* Receive response. */ 413 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) { 414 pjdlog_errno(LOG_ERR, "Unable to receive reload reply"); 415 return; 416 } 417 error = nv_get_int16(nvin, "error"); 418 nv_free(nvin); 419 if (error != 0) { 420 pjdlog_common(LOG_ERR, 0, error, "Reload failed"); 421 return; 422 } 423} 424 425static void 426hastd_reload(void) 427{ 428 struct hastd_config *newcfg; 429 struct hast_resource *nres, *cres, *tres; 430 uint8_t role; 431 432 pjdlog_info("Reloading configuration..."); 433 434 newcfg = yy_config_parse(cfgpath, false); 435 if (newcfg == NULL) 436 goto failed; 437 438 /* 439 * Check if control address has changed. 440 */ 441 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 442 if (proto_server(newcfg->hc_controladdr, 443 &newcfg->hc_controlconn) < 0) { 444 pjdlog_errno(LOG_ERR, 445 "Unable to listen on control address %s", 446 newcfg->hc_controladdr); 447 goto failed; 448 } 449 } 450 /* 451 * Check if listen address has changed. 452 */ 453 if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) { 454 if (proto_server(newcfg->hc_listenaddr, 455 &newcfg->hc_listenconn) < 0) { 456 pjdlog_errno(LOG_ERR, "Unable to listen on address %s", 457 newcfg->hc_listenaddr); 458 goto failed; 459 } 460 } 461 /* 462 * Only when both control and listen sockets are successfully 463 * initialized switch them to new configuration. 464 */ 465 if (newcfg->hc_controlconn != NULL) { 466 pjdlog_info("Control socket changed from %s to %s.", 467 cfg->hc_controladdr, newcfg->hc_controladdr); 468 proto_close(cfg->hc_controlconn); 469 cfg->hc_controlconn = newcfg->hc_controlconn; 470 newcfg->hc_controlconn = NULL; 471 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 472 sizeof(cfg->hc_controladdr)); 473 } 474 if (newcfg->hc_listenconn != NULL) { 475 pjdlog_info("Listen socket changed from %s to %s.", 476 cfg->hc_listenaddr, newcfg->hc_listenaddr); 477 proto_close(cfg->hc_listenconn); 478 cfg->hc_listenconn = newcfg->hc_listenconn; 479 newcfg->hc_listenconn = NULL; 480 strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr, 481 sizeof(cfg->hc_listenaddr)); 482 } 483 484 /* 485 * Stop and remove resources that were removed from the configuration. 486 */ 487 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 488 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 489 if (strcmp(cres->hr_name, nres->hr_name) == 0) 490 break; 491 } 492 if (nres == NULL) { 493 control_set_role(cres, HAST_ROLE_INIT); 494 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 495 pjdlog_info("Resource %s removed.", cres->hr_name); 496 free(cres); 497 } 498 } 499 /* 500 * Move new resources to the current configuration. 501 */ 502 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 503 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 504 if (strcmp(cres->hr_name, nres->hr_name) == 0) 505 break; 506 } 507 if (cres == NULL) { 508 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 509 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 510 pjdlog_info("Resource %s added.", nres->hr_name); 511 } 512 } 513 /* 514 * Deal with modified resources. 515 * Depending on what has changed exactly we might want to perform 516 * different actions. 517 * 518 * We do full resource restart in the following situations: 519 * Resource role is INIT or SECONDARY. 520 * Resource role is PRIMARY and path to local component or provider 521 * name has changed. 522 * In case of PRIMARY, the worker process will be killed and restarted, 523 * which also means removing /dev/hast/<name> provider and 524 * recreating it. 525 * 526 * We do just reload (send SIGHUP to worker process) if we act as 527 * PRIMARY, but only if remote address, replication mode, timeout or 528 * execution path has changed. For those, there is no need to restart 529 * worker process. 530 * If PRIMARY receives SIGHUP, it will reconnect if remote address or 531 * replication mode has changed or simply set new timeout if only 532 * timeout has changed. 533 */ 534 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 535 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 536 if (strcmp(cres->hr_name, nres->hr_name) == 0) 537 break; 538 } 539 PJDLOG_ASSERT(cres != NULL); 540 if (resource_needs_restart(cres, nres)) { 541 pjdlog_info("Resource %s configuration was modified, restarting it.", 542 cres->hr_name); 543 role = cres->hr_role; 544 control_set_role(cres, HAST_ROLE_INIT); 545 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 546 free(cres); 547 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 548 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 549 control_set_role(nres, role); 550 } else if (resource_needs_reload(cres, nres)) { 551 pjdlog_info("Resource %s configuration was modified, reloading it.", 552 cres->hr_name); 553 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 554 sizeof(cres->hr_remoteaddr)); 555 cres->hr_replication = nres->hr_replication; 556 cres->hr_timeout = nres->hr_timeout; 557 strlcpy(cres->hr_exec, nres->hr_exec, 558 sizeof(cres->hr_exec)); 559 if (cres->hr_workerpid != 0) 560 resource_reload(cres); 561 } 562 } 563 564 yy_config_free(newcfg); 565 pjdlog_info("Configuration reloaded successfully."); 566 return; 567failed: 568 if (newcfg != NULL) { 569 if (newcfg->hc_controlconn != NULL) 570 proto_close(newcfg->hc_controlconn); 571 if (newcfg->hc_listenconn != NULL) 572 proto_close(newcfg->hc_listenconn); 573 yy_config_free(newcfg); 574 } 575 pjdlog_warning("Configuration not reloaded."); 576} 577 578static void 579terminate_workers(void) 580{ 581 struct hast_resource *res; 582 583 pjdlog_info("Termination signal received, exiting."); 584 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 585 if (res->hr_workerpid == 0) 586 continue; 587 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 588 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 589 if (kill(res->hr_workerpid, SIGTERM) == 0) 590 continue; 591 pjdlog_errno(LOG_WARNING, 592 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 593 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 594 } 595} 596 597static void 598listen_accept(void) 599{ 600 struct hast_resource *res; 601 struct proto_conn *conn; 602 struct nv *nvin, *nvout, *nverr; 603 const char *resname; 604 const unsigned char *token; 605 char laddr[256], raddr[256]; 606 size_t size; 607 pid_t pid; 608 int status; 609 610 proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr)); 611 pjdlog_debug(1, "Accepting connection to %s.", laddr); 612 613 if (proto_accept(cfg->hc_listenconn, &conn) < 0) { 614 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 615 return; 616 } 617 618 proto_local_address(conn, laddr, sizeof(laddr)); 619 proto_remote_address(conn, raddr, sizeof(raddr)); 620 pjdlog_info("Connection from %s to %s.", raddr, laddr); 621 622 /* Error in setting timeout is not critical, but why should it fail? */ 623 if (proto_timeout(conn, HAST_TIMEOUT) < 0) 624 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 625 626 nvin = nvout = nverr = NULL; 627 628 /* 629 * Before receiving any data see if remote host have access to any 630 * resource. 631 */ 632 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 633 if (proto_address_match(conn, res->hr_remoteaddr)) 634 break; 635 } 636 if (res == NULL) { 637 pjdlog_error("Client %s isn't known.", raddr); 638 goto close; 639 } 640 /* Ok, remote host can access at least one resource. */ 641 642 if (hast_proto_recv_hdr(conn, &nvin) < 0) { 643 pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 644 raddr); 645 goto close; 646 } 647 648 resname = nv_get_string(nvin, "resource"); 649 if (resname == NULL) { 650 pjdlog_error("No 'resource' field in the header received from %s.", 651 raddr); 652 goto close; 653 } 654 pjdlog_debug(2, "%s: resource=%s", raddr, resname); 655 token = nv_get_uint8_array(nvin, &size, "token"); 656 /* 657 * NULL token means that this is first conection. 658 */ 659 if (token != NULL && size != sizeof(res->hr_token)) { 660 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 661 raddr, sizeof(res->hr_token), size); 662 goto close; 663 } 664 665 /* 666 * From now on we want to send errors to the remote node. 667 */ 668 nverr = nv_alloc(); 669 670 /* Find resource related to this connection. */ 671 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 672 if (strcmp(resname, res->hr_name) == 0) 673 break; 674 } 675 /* Have we found the resource? */ 676 if (res == NULL) { 677 pjdlog_error("No resource '%s' as requested by %s.", 678 resname, raddr); 679 nv_add_stringf(nverr, "errmsg", "Resource not configured."); 680 goto fail; 681 } 682 683 /* Now that we know resource name setup log prefix. */ 684 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 685 686 /* Does the remote host have access to this resource? */ 687 if (!proto_address_match(conn, res->hr_remoteaddr)) { 688 pjdlog_error("Client %s has no access to the resource.", raddr); 689 nv_add_stringf(nverr, "errmsg", "No access to the resource."); 690 goto fail; 691 } 692 /* Is the resource marked as secondary? */ 693 if (res->hr_role != HAST_ROLE_SECONDARY) { 694 pjdlog_error("We act as %s for the resource and not as %s as requested by %s.", 695 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 696 raddr); 697 nv_add_stringf(nverr, "errmsg", 698 "Remote node acts as %s for the resource and not as %s.", 699 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 700 goto fail; 701 } 702 /* Does token (if exists) match? */ 703 if (token != NULL && memcmp(token, res->hr_token, 704 sizeof(res->hr_token)) != 0) { 705 pjdlog_error("Token received from %s doesn't match.", raddr); 706 nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 707 goto fail; 708 } 709 /* 710 * If there is no token, but we have half-open connection 711 * (only remotein) or full connection (worker process is running) 712 * we have to cancel those and accept the new connection. 713 */ 714 if (token == NULL) { 715 PJDLOG_ASSERT(res->hr_remoteout == NULL); 716 pjdlog_debug(1, "Initial connection from %s.", raddr); 717 if (res->hr_workerpid != 0) { 718 PJDLOG_ASSERT(res->hr_remotein == NULL); 719 pjdlog_debug(1, 720 "Worker process exists (pid=%u), stopping it.", 721 (unsigned int)res->hr_workerpid); 722 /* Stop child process. */ 723 if (kill(res->hr_workerpid, SIGINT) < 0) { 724 pjdlog_errno(LOG_ERR, 725 "Unable to stop worker process (pid=%u)", 726 (unsigned int)res->hr_workerpid); 727 /* 728 * Other than logging the problem we 729 * ignore it - nothing smart to do. 730 */ 731 } 732 /* Wait for it to exit. */ 733 else if ((pid = waitpid(res->hr_workerpid, 734 &status, 0)) != res->hr_workerpid) { 735 /* We can only log the problem. */ 736 pjdlog_errno(LOG_ERR, 737 "Waiting for worker process (pid=%u) failed", 738 (unsigned int)res->hr_workerpid); 739 } else { 740 child_exit_log(res->hr_workerpid, status); 741 } 742 child_cleanup(res); 743 } else if (res->hr_remotein != NULL) { 744 char oaddr[256]; 745 746 proto_remote_address(res->hr_remotein, oaddr, 747 sizeof(oaddr)); 748 pjdlog_debug(1, 749 "Canceling half-open connection from %s on connection from %s.", 750 oaddr, raddr); 751 proto_close(res->hr_remotein); 752 res->hr_remotein = NULL; 753 } 754 } 755 756 /* 757 * Checks and cleanups are done. 758 */ 759 760 if (token == NULL) { 761 arc4random_buf(res->hr_token, sizeof(res->hr_token)); 762 nvout = nv_alloc(); 763 nv_add_uint8_array(nvout, res->hr_token, 764 sizeof(res->hr_token), "token"); 765 if (nv_error(nvout) != 0) { 766 pjdlog_common(LOG_ERR, 0, nv_error(nvout), 767 "Unable to prepare return header for %s", raddr); 768 nv_add_stringf(nverr, "errmsg", 769 "Remote node was unable to prepare return header: %s.", 770 strerror(nv_error(nvout))); 771 goto fail; 772 } 773 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) { 774 int error = errno; 775 776 pjdlog_errno(LOG_ERR, "Unable to send response to %s", 777 raddr); 778 nv_add_stringf(nverr, "errmsg", 779 "Remote node was unable to send response: %s.", 780 strerror(error)); 781 goto fail; 782 } 783 res->hr_remotein = conn; 784 pjdlog_debug(1, "Incoming connection from %s configured.", 785 raddr); 786 } else { 787 res->hr_remoteout = conn; 788 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 789 hastd_secondary(res, nvin); 790 } 791 nv_free(nvin); 792 nv_free(nvout); 793 nv_free(nverr); 794 pjdlog_prefix_set("%s", ""); 795 return; 796fail: 797 if (nv_error(nverr) != 0) { 798 pjdlog_common(LOG_ERR, 0, nv_error(nverr), 799 "Unable to prepare error header for %s", raddr); 800 goto close; 801 } 802 if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) { 803 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 804 goto close; 805 } 806close: 807 if (nvin != NULL) 808 nv_free(nvin); 809 if (nvout != NULL) 810 nv_free(nvout); 811 if (nverr != NULL) 812 nv_free(nverr); 813 proto_close(conn); 814 pjdlog_prefix_set("%s", ""); 815} 816 817static void 818connection_migrate(struct hast_resource *res) 819{ 820 struct proto_conn *conn; 821 int16_t val = 0; 822 823 if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) { 824 pjdlog_errno(LOG_WARNING, 825 "Unable to receive connection command"); 826 return; 827 } 828 if (proto_client(res->hr_remoteaddr, &conn) < 0) { 829 val = errno; 830 pjdlog_errno(LOG_WARNING, 831 "Unable to create outgoing connection to %s", 832 res->hr_remoteaddr); 833 goto out; 834 } 835 if (proto_connect(conn, -1) < 0) { 836 val = errno; 837 pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 838 res->hr_remoteaddr); 839 proto_close(conn); 840 goto out; 841 } 842 val = 0; 843out: 844 if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) { 845 pjdlog_errno(LOG_WARNING, 846 "Unable to send reply to connection request"); 847 } 848 if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0) 849 pjdlog_errno(LOG_WARNING, "Unable to send connection"); 850} 851 852static void 853main_loop(void) 854{ 855 struct hast_resource *res; 856 struct timeval seltimeout; 857 struct timespec sigtimeout; 858 int fd, maxfd, ret, signo; 859 sigset_t mask; 860 fd_set rfds; 861 862 seltimeout.tv_sec = REPORT_INTERVAL; 863 seltimeout.tv_usec = 0; 864 sigtimeout.tv_sec = 0; 865 sigtimeout.tv_nsec = 0; 866 867 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 868 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 869 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 870 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 871 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 872 873 pjdlog_info("Started successfully, running protocol version %d.", 874 HAST_PROTO_VERSION); 875 876 for (;;) { 877 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 878 switch (signo) { 879 case SIGINT: 880 case SIGTERM: 881 sigexit_received = true; 882 terminate_workers(); 883 proto_close(cfg->hc_controlconn); 884 exit(EX_OK); 885 break; 886 case SIGCHLD: 887 child_exit(); 888 break; 889 case SIGHUP: 890 hastd_reload(); 891 break; 892 default: 893 PJDLOG_ABORT("Unexpected signal (%d).", signo); 894 } 895 } 896 897 /* Setup descriptors for select(2). */ 898 FD_ZERO(&rfds); 899 maxfd = fd = proto_descriptor(cfg->hc_controlconn); 900 PJDLOG_ASSERT(fd >= 0); 901 FD_SET(fd, &rfds); 902 fd = proto_descriptor(cfg->hc_listenconn); 903 PJDLOG_ASSERT(fd >= 0); 904 FD_SET(fd, &rfds); 905 maxfd = fd > maxfd ? fd : maxfd; 906 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 907 if (res->hr_event == NULL) 908 continue; 909 PJDLOG_ASSERT(res->hr_conn != NULL); 910 fd = proto_descriptor(res->hr_event); 911 PJDLOG_ASSERT(fd >= 0); 912 FD_SET(fd, &rfds); 913 maxfd = fd > maxfd ? fd : maxfd; 914 if (res->hr_role == HAST_ROLE_PRIMARY) { 915 /* Only primary workers asks for connections. */ 916 fd = proto_descriptor(res->hr_conn); 917 PJDLOG_ASSERT(fd >= 0); 918 FD_SET(fd, &rfds); 919 maxfd = fd > maxfd ? fd : maxfd; 920 } 921 } 922 923 PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE); 924 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 925 if (ret == 0) 926 hook_check(); 927 else if (ret == -1) { 928 if (errno == EINTR) 929 continue; 930 KEEP_ERRNO((void)pidfile_remove(pfh)); 931 pjdlog_exit(EX_OSERR, "select() failed"); 932 } 933 934 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 935 control_handle(cfg); 936 if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds)) 937 listen_accept(); 938 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 939 if (res->hr_event == NULL) 940 continue; 941 PJDLOG_ASSERT(res->hr_conn != NULL); 942 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 943 if (event_recv(res) == 0) 944 continue; 945 /* The worker process exited? */ 946 proto_close(res->hr_event); 947 res->hr_event = NULL; 948 proto_close(res->hr_conn); 949 res->hr_conn = NULL; 950 continue; 951 } 952 if (res->hr_role == HAST_ROLE_PRIMARY && 953 FD_ISSET(proto_descriptor(res->hr_conn), &rfds)) { 954 connection_migrate(res); 955 } 956 } 957 } 958} 959 960static void 961dummy_sighandler(int sig __unused) 962{ 963 /* Nothing to do. */ 964} 965 966int 967main(int argc, char *argv[]) 968{ 969 const char *pidfile; 970 pid_t otherpid; 971 bool foreground; 972 int debuglevel; 973 sigset_t mask; 974 975 foreground = false; 976 debuglevel = 0; 977 pidfile = HASTD_PIDFILE; 978 979 for (;;) { 980 int ch; 981 982 ch = getopt(argc, argv, "c:dFhP:"); 983 if (ch == -1) 984 break; 985 switch (ch) { 986 case 'c': 987 cfgpath = optarg; 988 break; 989 case 'd': 990 debuglevel++; 991 break; 992 case 'F': 993 foreground = true; 994 break; 995 case 'P': 996 pidfile = optarg; 997 break; 998 case 'h': 999 default: 1000 usage(); 1001 } 1002 } 1003 argc -= optind; 1004 argv += optind; 1005 1006 pjdlog_init(PJDLOG_MODE_STD); 1007 pjdlog_debug_set(debuglevel); 1008 1009 g_gate_load(); 1010 1011 pfh = pidfile_open(pidfile, 0600, &otherpid); 1012 if (pfh == NULL) { 1013 if (errno == EEXIST) { 1014 pjdlog_exitx(EX_TEMPFAIL, 1015 "Another hastd is already running, pid: %jd.", 1016 (intmax_t)otherpid); 1017 } 1018 /* If we cannot create pidfile from other reasons, only warn. */ 1019 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile"); 1020 } 1021 1022 cfg = yy_config_parse(cfgpath, true); 1023 PJDLOG_ASSERT(cfg != NULL); 1024 1025 /* 1026 * Restore default actions for interesting signals in case parent 1027 * process (like init(8)) decided to ignore some of them (like SIGHUP). 1028 */ 1029 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 1030 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 1031 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 1032 /* 1033 * Because SIGCHLD is ignored by default, setup dummy handler for it, 1034 * so we can mask it. 1035 */ 1036 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 1037 1038 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1039 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1040 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1041 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1042 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1043 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 1044 1045 /* Listen on control address. */ 1046 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) { 1047 KEEP_ERRNO((void)pidfile_remove(pfh)); 1048 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 1049 cfg->hc_controladdr); 1050 } 1051 /* Listen for remote connections. */ 1052 if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) { 1053 KEEP_ERRNO((void)pidfile_remove(pfh)); 1054 pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 1055 cfg->hc_listenaddr); 1056 } 1057 1058 if (!foreground) { 1059 if (daemon(0, 0) < 0) { 1060 KEEP_ERRNO((void)pidfile_remove(pfh)); 1061 pjdlog_exit(EX_OSERR, "Unable to daemonize"); 1062 } 1063 1064 /* Start logging to syslog. */ 1065 pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 1066 1067 /* Write PID to a file. */ 1068 if (pidfile_write(pfh) < 0) { 1069 pjdlog_errno(LOG_WARNING, 1070 "Unable to write PID to a file"); 1071 } 1072 } 1073 1074 hook_init(); 1075 1076 main_loop(); 1077 1078 exit(0); 1079} 1080