hastd.c revision 218370
1130812Smarcel/*- 2130812Smarcel * Copyright (c) 2009-2010 The FreeBSD Foundation 3130812Smarcel * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pjd@FreeBSD.org> 4130812Smarcel * All rights reserved. 5130812Smarcel * 6130812Smarcel * This software was developed by Pawel Jakub Dawidek under sponsorship from 7130812Smarcel * the FreeBSD Foundation. 8130812Smarcel * 9130812Smarcel * Redistribution and use in source and binary forms, with or without 10130812Smarcel * modification, are permitted provided that the following conditions 11130812Smarcel * are met: 12130812Smarcel * 1. Redistributions of source code must retain the above copyright 13130812Smarcel * notice, this list of conditions and the following disclaimer. 14130812Smarcel * 2. Redistributions in binary form must reproduce the above copyright 15130812Smarcel * notice, this list of conditions and the following disclaimer in the 16130812Smarcel * documentation and/or other materials provided with the distribution. 17130812Smarcel * 18130812Smarcel * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 19130812Smarcel * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20130812Smarcel * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21130812Smarcel * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 22130812Smarcel * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23130812Smarcel * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24130812Smarcel * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25130812Smarcel * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26130812Smarcel * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27130812Smarcel * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28130812Smarcel * SUCH DAMAGE. 29130812Smarcel */ 30130812Smarcel 31130812Smarcel#include <sys/cdefs.h> 32130812Smarcel__FBSDID("$FreeBSD: head/sbin/hastd/hastd.c 218370 2011-02-06 12:21:29Z pjd $"); 33130812Smarcel 34130812Smarcel#include <sys/param.h> 35130812Smarcel#include <sys/linker.h> 36130812Smarcel#include <sys/module.h> 37130812Smarcel#include <sys/stat.h> 38130812Smarcel#include <sys/wait.h> 39130812Smarcel 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 if (tres->hr_ctrl != NULL) 113 proto_close(tres->hr_ctrl); 114 if (tres->hr_event != NULL) 115 proto_close(tres->hr_event); 116 if (tres->hr_conn != NULL) 117 proto_close(tres->hr_conn); 118 } 119 if (cfg->hc_controlin != NULL) 120 proto_close(cfg->hc_controlin); 121 proto_close(cfg->hc_controlconn); 122 proto_close(cfg->hc_listenconn); 123 (void)pidfile_close(pfh); 124 hook_fini(); 125 pjdlog_fini(); 126} 127 128static const char * 129dtype2str(mode_t mode) 130{ 131 132 if (S_ISBLK(mode)) 133 return ("block device"); 134 else if (S_ISCHR(mode)) 135 return ("character device"); 136 else if (S_ISDIR(mode)) 137 return ("directory"); 138 else if (S_ISFIFO(mode)) 139 return ("pipe or FIFO"); 140 else if (S_ISLNK(mode)) 141 return ("symbolic link"); 142 else if (S_ISREG(mode)) 143 return ("regular file"); 144 else if (S_ISSOCK(mode)) 145 return ("socket"); 146 else if (S_ISWHT(mode)) 147 return ("whiteout"); 148 else 149 return ("unknown"); 150} 151 152void 153descriptors_assert(const struct hast_resource *res, int pjdlogmode) 154{ 155 char msg[256]; 156 struct stat sb; 157 long maxfd; 158 bool isopen; 159 mode_t mode; 160 int fd; 161 162 /* 163 * At this point descriptor to syslog socket is closed, so if we want 164 * to log assertion message, we have to first store it in 'msg' local 165 * buffer and then open syslog socket and log it. 166 */ 167 msg[0] = '\0'; 168 169 maxfd = sysconf(_SC_OPEN_MAX); 170 if (maxfd < 0) { 171 pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed"); 172 maxfd = 16384; 173 } 174 for (fd = 0; fd <= maxfd; fd++) { 175 if (fstat(fd, &sb) == 0) { 176 isopen = true; 177 mode = sb.st_mode; 178 } else if (errno == EBADF) { 179 isopen = false; 180 mode = 0; 181 } else { 182 isopen = true; /* silence gcc */ 183 mode = 0; /* silence gcc */ 184 snprintf(msg, sizeof(msg), 185 "Unable to fstat descriptor %d: %s", fd, 186 strerror(errno)); 187 } 188 if (fd == STDIN_FILENO || fd == STDOUT_FILENO || 189 fd == STDERR_FILENO) { 190 if (!isopen) { 191 snprintf(msg, sizeof(msg), 192 "Descriptor %d (%s) is closed, but should be open.", 193 fd, (fd == STDIN_FILENO ? "stdin" : 194 (fd == STDOUT_FILENO ? "stdout" : "stderr"))); 195 break; 196 } 197 } else if (fd == proto_descriptor(res->hr_event)) { 198 if (!isopen) { 199 snprintf(msg, sizeof(msg), 200 "Descriptor %d (event) is closed, but should be open.", 201 fd); 202 break; 203 } 204 if (!S_ISSOCK(mode)) { 205 snprintf(msg, sizeof(msg), 206 "Descriptor %d (event) is %s, but should be %s.", 207 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 208 break; 209 } 210 } else if (fd == proto_descriptor(res->hr_ctrl)) { 211 if (!isopen) { 212 snprintf(msg, sizeof(msg), 213 "Descriptor %d (ctrl) is closed, but should be open.", 214 fd); 215 break; 216 } 217 if (!S_ISSOCK(mode)) { 218 snprintf(msg, sizeof(msg), 219 "Descriptor %d (ctrl) is %s, but should be %s.", 220 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 221 break; 222 } 223 } else if (fd == proto_descriptor(res->hr_conn)) { 224 if (!isopen) { 225 snprintf(msg, sizeof(msg), 226 "Descriptor %d (conn) is closed, but should be open.", 227 fd); 228 break; 229 } 230 if (!S_ISSOCK(mode)) { 231 snprintf(msg, sizeof(msg), 232 "Descriptor %d (conn) is %s, but should be %s.", 233 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 234 break; 235 } 236 } else if (res->hr_role == HAST_ROLE_SECONDARY && 237 fd == proto_descriptor(res->hr_remotein)) { 238 if (!isopen) { 239 snprintf(msg, sizeof(msg), 240 "Descriptor %d (remote in) is closed, but should be open.", 241 fd); 242 break; 243 } 244 if (!S_ISSOCK(mode)) { 245 snprintf(msg, sizeof(msg), 246 "Descriptor %d (remote in) is %s, but should be %s.", 247 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 248 break; 249 } 250 } else if (res->hr_role == HAST_ROLE_SECONDARY && 251 fd == proto_descriptor(res->hr_remoteout)) { 252 if (!isopen) { 253 snprintf(msg, sizeof(msg), 254 "Descriptor %d (remote out) is closed, but should be open.", 255 fd); 256 break; 257 } 258 if (!S_ISSOCK(mode)) { 259 snprintf(msg, sizeof(msg), 260 "Descriptor %d (remote out) is %s, but should be %s.", 261 fd, dtype2str(mode), dtype2str(S_IFSOCK)); 262 break; 263 } 264 } else { 265 if (isopen) { 266 snprintf(msg, sizeof(msg), 267 "Descriptor %d is open (%s), but should be closed.", 268 fd, dtype2str(mode)); 269 break; 270 } 271 } 272 } 273 if (msg[0] != '\0') { 274 pjdlog_init(pjdlogmode); 275 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 276 role2str(res->hr_role)); 277 PJDLOG_ABORT("%s", msg); 278 } 279} 280 281static void 282child_exit_log(unsigned int pid, int status) 283{ 284 285 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 286 pjdlog_debug(1, "Worker process exited gracefully (pid=%u).", 287 pid); 288 } else if (WIFSIGNALED(status)) { 289 pjdlog_error("Worker process killed (pid=%u, signal=%d).", 290 pid, WTERMSIG(status)); 291 } else { 292 pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).", 293 pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1); 294 } 295} 296 297static void 298child_exit(void) 299{ 300 struct hast_resource *res; 301 int status; 302 pid_t pid; 303 304 while ((pid = wait3(&status, WNOHANG, NULL)) > 0) { 305 /* Find resource related to the process that just exited. */ 306 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 307 if (pid == res->hr_workerpid) 308 break; 309 } 310 if (res == NULL) { 311 /* 312 * This can happen when new connection arrives and we 313 * cancel child responsible for the old one or if this 314 * was hook which we executed. 315 */ 316 hook_check_one(pid, status); 317 continue; 318 } 319 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 320 role2str(res->hr_role)); 321 child_exit_log(pid, status); 322 child_cleanup(res); 323 if (res->hr_role == HAST_ROLE_PRIMARY) { 324 /* 325 * Restart child process if it was killed by signal 326 * or exited because of temporary problem. 327 */ 328 if (WIFSIGNALED(status) || 329 (WIFEXITED(status) && 330 WEXITSTATUS(status) == EX_TEMPFAIL)) { 331 sleep(1); 332 pjdlog_info("Restarting worker process."); 333 hastd_primary(res); 334 } else { 335 res->hr_role = HAST_ROLE_INIT; 336 pjdlog_info("Changing resource role back to %s.", 337 role2str(res->hr_role)); 338 } 339 } 340 pjdlog_prefix_set("%s", ""); 341 } 342} 343 344static bool 345resource_needs_restart(const struct hast_resource *res0, 346 const struct hast_resource *res1) 347{ 348 349 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 350 351 if (strcmp(res0->hr_provname, res1->hr_provname) != 0) 352 return (true); 353 if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0) 354 return (true); 355 if (res0->hr_role == HAST_ROLE_INIT || 356 res0->hr_role == HAST_ROLE_SECONDARY) { 357 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 358 return (true); 359 if (res0->hr_replication != res1->hr_replication) 360 return (true); 361 if (res0->hr_timeout != res1->hr_timeout) 362 return (true); 363 if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 364 return (true); 365 } 366 return (false); 367} 368 369static bool 370resource_needs_reload(const struct hast_resource *res0, 371 const struct hast_resource *res1) 372{ 373 374 PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 375 PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0); 376 PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0); 377 378 if (res0->hr_role != HAST_ROLE_PRIMARY) 379 return (false); 380 381 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 382 return (true); 383 if (res0->hr_replication != res1->hr_replication) 384 return (true); 385 if (res0->hr_timeout != res1->hr_timeout) 386 return (true); 387 if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 388 return (true); 389 return (false); 390} 391 392static void 393resource_reload(const struct hast_resource *res) 394{ 395 struct nv *nvin, *nvout; 396 int error; 397 398 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 399 400 nvout = nv_alloc(); 401 nv_add_uint8(nvout, HASTCTL_RELOAD, "cmd"); 402 nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr"); 403 nv_add_int32(nvout, (int32_t)res->hr_replication, "replication"); 404 nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout"); 405 nv_add_string(nvout, res->hr_exec, "exec"); 406 if (nv_error(nvout) != 0) { 407 nv_free(nvout); 408 pjdlog_error("Unable to allocate header for reload message."); 409 return; 410 } 411 if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) { 412 pjdlog_errno(LOG_ERR, "Unable to send reload message"); 413 nv_free(nvout); 414 return; 415 } 416 nv_free(nvout); 417 418 /* Receive response. */ 419 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) { 420 pjdlog_errno(LOG_ERR, "Unable to receive reload reply"); 421 return; 422 } 423 error = nv_get_int16(nvin, "error"); 424 nv_free(nvin); 425 if (error != 0) { 426 pjdlog_common(LOG_ERR, 0, error, "Reload failed"); 427 return; 428 } 429} 430 431static void 432hastd_reload(void) 433{ 434 struct hastd_config *newcfg; 435 struct hast_resource *nres, *cres, *tres; 436 uint8_t role; 437 438 pjdlog_info("Reloading configuration..."); 439 440 newcfg = yy_config_parse(cfgpath, false); 441 if (newcfg == NULL) 442 goto failed; 443 444 /* 445 * Check if control address has changed. 446 */ 447 if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 448 if (proto_server(newcfg->hc_controladdr, 449 &newcfg->hc_controlconn) < 0) { 450 pjdlog_errno(LOG_ERR, 451 "Unable to listen on control address %s", 452 newcfg->hc_controladdr); 453 goto failed; 454 } 455 } 456 /* 457 * Check if listen address has changed. 458 */ 459 if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) { 460 if (proto_server(newcfg->hc_listenaddr, 461 &newcfg->hc_listenconn) < 0) { 462 pjdlog_errno(LOG_ERR, "Unable to listen on address %s", 463 newcfg->hc_listenaddr); 464 goto failed; 465 } 466 } 467 /* 468 * Only when both control and listen sockets are successfully 469 * initialized switch them to new configuration. 470 */ 471 if (newcfg->hc_controlconn != NULL) { 472 pjdlog_info("Control socket changed from %s to %s.", 473 cfg->hc_controladdr, newcfg->hc_controladdr); 474 proto_close(cfg->hc_controlconn); 475 cfg->hc_controlconn = newcfg->hc_controlconn; 476 newcfg->hc_controlconn = NULL; 477 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 478 sizeof(cfg->hc_controladdr)); 479 } 480 if (newcfg->hc_listenconn != NULL) { 481 pjdlog_info("Listen socket changed from %s to %s.", 482 cfg->hc_listenaddr, newcfg->hc_listenaddr); 483 proto_close(cfg->hc_listenconn); 484 cfg->hc_listenconn = newcfg->hc_listenconn; 485 newcfg->hc_listenconn = NULL; 486 strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr, 487 sizeof(cfg->hc_listenaddr)); 488 } 489 490 /* 491 * Stop and remove resources that were removed from the configuration. 492 */ 493 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 494 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 495 if (strcmp(cres->hr_name, nres->hr_name) == 0) 496 break; 497 } 498 if (nres == NULL) { 499 control_set_role(cres, HAST_ROLE_INIT); 500 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 501 pjdlog_info("Resource %s removed.", cres->hr_name); 502 free(cres); 503 } 504 } 505 /* 506 * Move new resources to the current configuration. 507 */ 508 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 509 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 510 if (strcmp(cres->hr_name, nres->hr_name) == 0) 511 break; 512 } 513 if (cres == NULL) { 514 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 515 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 516 pjdlog_info("Resource %s added.", nres->hr_name); 517 } 518 } 519 /* 520 * Deal with modified resources. 521 * Depending on what has changed exactly we might want to perform 522 * different actions. 523 * 524 * We do full resource restart in the following situations: 525 * Resource role is INIT or SECONDARY. 526 * Resource role is PRIMARY and path to local component or provider 527 * name has changed. 528 * In case of PRIMARY, the worker process will be killed and restarted, 529 * which also means removing /dev/hast/<name> provider and 530 * recreating it. 531 * 532 * We do just reload (send SIGHUP to worker process) if we act as 533 * PRIMARY, but only if remote address, replication mode, timeout or 534 * execution path has changed. For those, there is no need to restart 535 * worker process. 536 * If PRIMARY receives SIGHUP, it will reconnect if remote address or 537 * replication mode has changed or simply set new timeout if only 538 * timeout has changed. 539 */ 540 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 541 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 542 if (strcmp(cres->hr_name, nres->hr_name) == 0) 543 break; 544 } 545 PJDLOG_ASSERT(cres != NULL); 546 if (resource_needs_restart(cres, nres)) { 547 pjdlog_info("Resource %s configuration was modified, restarting it.", 548 cres->hr_name); 549 role = cres->hr_role; 550 control_set_role(cres, HAST_ROLE_INIT); 551 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 552 free(cres); 553 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 554 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 555 control_set_role(nres, role); 556 } else if (resource_needs_reload(cres, nres)) { 557 pjdlog_info("Resource %s configuration was modified, reloading it.", 558 cres->hr_name); 559 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 560 sizeof(cres->hr_remoteaddr)); 561 cres->hr_replication = nres->hr_replication; 562 cres->hr_timeout = nres->hr_timeout; 563 strlcpy(cres->hr_exec, nres->hr_exec, 564 sizeof(cres->hr_exec)); 565 if (cres->hr_workerpid != 0) 566 resource_reload(cres); 567 } 568 } 569 570 yy_config_free(newcfg); 571 pjdlog_info("Configuration reloaded successfully."); 572 return; 573failed: 574 if (newcfg != NULL) { 575 if (newcfg->hc_controlconn != NULL) 576 proto_close(newcfg->hc_controlconn); 577 if (newcfg->hc_listenconn != NULL) 578 proto_close(newcfg->hc_listenconn); 579 yy_config_free(newcfg); 580 } 581 pjdlog_warning("Configuration not reloaded."); 582} 583 584static void 585terminate_workers(void) 586{ 587 struct hast_resource *res; 588 589 pjdlog_info("Termination signal received, exiting."); 590 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 591 if (res->hr_workerpid == 0) 592 continue; 593 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 594 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 595 if (kill(res->hr_workerpid, SIGTERM) == 0) 596 continue; 597 pjdlog_errno(LOG_WARNING, 598 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 599 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 600 } 601} 602 603static void 604listen_accept(void) 605{ 606 struct hast_resource *res; 607 struct proto_conn *conn; 608 struct nv *nvin, *nvout, *nverr; 609 const char *resname; 610 const unsigned char *token; 611 char laddr[256], raddr[256]; 612 size_t size; 613 pid_t pid; 614 int status; 615 616 proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr)); 617 pjdlog_debug(1, "Accepting connection to %s.", laddr); 618 619 if (proto_accept(cfg->hc_listenconn, &conn) < 0) { 620 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 621 return; 622 } 623 624 proto_local_address(conn, laddr, sizeof(laddr)); 625 proto_remote_address(conn, raddr, sizeof(raddr)); 626 pjdlog_info("Connection from %s to %s.", raddr, laddr); 627 628 /* Error in setting timeout is not critical, but why should it fail? */ 629 if (proto_timeout(conn, HAST_TIMEOUT) < 0) 630 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 631 632 nvin = nvout = nverr = NULL; 633 634 /* 635 * Before receiving any data see if remote host have access to any 636 * resource. 637 */ 638 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 639 if (proto_address_match(conn, res->hr_remoteaddr)) 640 break; 641 } 642 if (res == NULL) { 643 pjdlog_error("Client %s isn't known.", raddr); 644 goto close; 645 } 646 /* Ok, remote host can access at least one resource. */ 647 648 if (hast_proto_recv_hdr(conn, &nvin) < 0) { 649 pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 650 raddr); 651 goto close; 652 } 653 654 resname = nv_get_string(nvin, "resource"); 655 if (resname == NULL) { 656 pjdlog_error("No 'resource' field in the header received from %s.", 657 raddr); 658 goto close; 659 } 660 pjdlog_debug(2, "%s: resource=%s", raddr, resname); 661 token = nv_get_uint8_array(nvin, &size, "token"); 662 /* 663 * NULL token means that this is first conection. 664 */ 665 if (token != NULL && size != sizeof(res->hr_token)) { 666 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 667 raddr, sizeof(res->hr_token), size); 668 goto close; 669 } 670 671 /* 672 * From now on we want to send errors to the remote node. 673 */ 674 nverr = nv_alloc(); 675 676 /* Find resource related to this connection. */ 677 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 678 if (strcmp(resname, res->hr_name) == 0) 679 break; 680 } 681 /* Have we found the resource? */ 682 if (res == NULL) { 683 pjdlog_error("No resource '%s' as requested by %s.", 684 resname, raddr); 685 nv_add_stringf(nverr, "errmsg", "Resource not configured."); 686 goto fail; 687 } 688 689 /* Now that we know resource name setup log prefix. */ 690 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 691 692 /* Does the remote host have access to this resource? */ 693 if (!proto_address_match(conn, res->hr_remoteaddr)) { 694 pjdlog_error("Client %s has no access to the resource.", raddr); 695 nv_add_stringf(nverr, "errmsg", "No access to the resource."); 696 goto fail; 697 } 698 /* Is the resource marked as secondary? */ 699 if (res->hr_role != HAST_ROLE_SECONDARY) { 700 pjdlog_error("We act as %s for the resource and not as %s as requested by %s.", 701 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 702 raddr); 703 nv_add_stringf(nverr, "errmsg", 704 "Remote node acts as %s for the resource and not as %s.", 705 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 706 goto fail; 707 } 708 /* Does token (if exists) match? */ 709 if (token != NULL && memcmp(token, res->hr_token, 710 sizeof(res->hr_token)) != 0) { 711 pjdlog_error("Token received from %s doesn't match.", raddr); 712 nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 713 goto fail; 714 } 715 /* 716 * If there is no token, but we have half-open connection 717 * (only remotein) or full connection (worker process is running) 718 * we have to cancel those and accept the new connection. 719 */ 720 if (token == NULL) { 721 PJDLOG_ASSERT(res->hr_remoteout == NULL); 722 pjdlog_debug(1, "Initial connection from %s.", raddr); 723 if (res->hr_workerpid != 0) { 724 PJDLOG_ASSERT(res->hr_remotein == NULL); 725 pjdlog_debug(1, 726 "Worker process exists (pid=%u), stopping it.", 727 (unsigned int)res->hr_workerpid); 728 /* Stop child process. */ 729 if (kill(res->hr_workerpid, SIGINT) < 0) { 730 pjdlog_errno(LOG_ERR, 731 "Unable to stop worker process (pid=%u)", 732 (unsigned int)res->hr_workerpid); 733 /* 734 * Other than logging the problem we 735 * ignore it - nothing smart to do. 736 */ 737 } 738 /* Wait for it to exit. */ 739 else if ((pid = waitpid(res->hr_workerpid, 740 &status, 0)) != res->hr_workerpid) { 741 /* We can only log the problem. */ 742 pjdlog_errno(LOG_ERR, 743 "Waiting for worker process (pid=%u) failed", 744 (unsigned int)res->hr_workerpid); 745 } else { 746 child_exit_log(res->hr_workerpid, status); 747 } 748 child_cleanup(res); 749 } else if (res->hr_remotein != NULL) { 750 char oaddr[256]; 751 752 proto_remote_address(res->hr_remotein, oaddr, 753 sizeof(oaddr)); 754 pjdlog_debug(1, 755 "Canceling half-open connection from %s on connection from %s.", 756 oaddr, raddr); 757 proto_close(res->hr_remotein); 758 res->hr_remotein = NULL; 759 } 760 } 761 762 /* 763 * Checks and cleanups are done. 764 */ 765 766 if (token == NULL) { 767 arc4random_buf(res->hr_token, sizeof(res->hr_token)); 768 nvout = nv_alloc(); 769 nv_add_uint8_array(nvout, res->hr_token, 770 sizeof(res->hr_token), "token"); 771 if (nv_error(nvout) != 0) { 772 pjdlog_common(LOG_ERR, 0, nv_error(nvout), 773 "Unable to prepare return header for %s", raddr); 774 nv_add_stringf(nverr, "errmsg", 775 "Remote node was unable to prepare return header: %s.", 776 strerror(nv_error(nvout))); 777 goto fail; 778 } 779 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) { 780 int error = errno; 781 782 pjdlog_errno(LOG_ERR, "Unable to send response to %s", 783 raddr); 784 nv_add_stringf(nverr, "errmsg", 785 "Remote node was unable to send response: %s.", 786 strerror(error)); 787 goto fail; 788 } 789 res->hr_remotein = conn; 790 pjdlog_debug(1, "Incoming connection from %s configured.", 791 raddr); 792 } else { 793 res->hr_remoteout = conn; 794 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 795 hastd_secondary(res, nvin); 796 } 797 nv_free(nvin); 798 nv_free(nvout); 799 nv_free(nverr); 800 pjdlog_prefix_set("%s", ""); 801 return; 802fail: 803 if (nv_error(nverr) != 0) { 804 pjdlog_common(LOG_ERR, 0, nv_error(nverr), 805 "Unable to prepare error header for %s", raddr); 806 goto close; 807 } 808 if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) { 809 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 810 goto close; 811 } 812close: 813 if (nvin != NULL) 814 nv_free(nvin); 815 if (nvout != NULL) 816 nv_free(nvout); 817 if (nverr != NULL) 818 nv_free(nverr); 819 proto_close(conn); 820 pjdlog_prefix_set("%s", ""); 821} 822 823static void 824connection_migrate(struct hast_resource *res) 825{ 826 struct proto_conn *conn; 827 int16_t val = 0; 828 829 if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) { 830 pjdlog_errno(LOG_WARNING, 831 "Unable to receive connection command"); 832 return; 833 } 834 if (proto_client(res->hr_remoteaddr, &conn) < 0) { 835 val = errno; 836 pjdlog_errno(LOG_WARNING, 837 "Unable to create outgoing connection to %s", 838 res->hr_remoteaddr); 839 goto out; 840 } 841 if (proto_connect(conn, -1) < 0) { 842 val = errno; 843 pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 844 res->hr_remoteaddr); 845 proto_close(conn); 846 goto out; 847 } 848 val = 0; 849out: 850 if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) { 851 pjdlog_errno(LOG_WARNING, 852 "Unable to send reply to connection request"); 853 } 854 if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0) 855 pjdlog_errno(LOG_WARNING, "Unable to send connection"); 856} 857 858static void 859main_loop(void) 860{ 861 struct hast_resource *res; 862 struct timeval seltimeout; 863 struct timespec sigtimeout; 864 int fd, maxfd, ret, signo; 865 sigset_t mask; 866 fd_set rfds; 867 868 seltimeout.tv_sec = REPORT_INTERVAL; 869 seltimeout.tv_usec = 0; 870 sigtimeout.tv_sec = 0; 871 sigtimeout.tv_nsec = 0; 872 873 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 874 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 875 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 876 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 877 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 878 879 pjdlog_info("Started successfully, running protocol version %d.", 880 HAST_PROTO_VERSION); 881 882 for (;;) { 883 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 884 switch (signo) { 885 case SIGINT: 886 case SIGTERM: 887 sigexit_received = true; 888 terminate_workers(); 889 proto_close(cfg->hc_controlconn); 890 exit(EX_OK); 891 break; 892 case SIGCHLD: 893 child_exit(); 894 break; 895 case SIGHUP: 896 hastd_reload(); 897 break; 898 default: 899 PJDLOG_ABORT("Unexpected signal (%d).", signo); 900 } 901 } 902 903 /* Setup descriptors for select(2). */ 904 FD_ZERO(&rfds); 905 maxfd = fd = proto_descriptor(cfg->hc_controlconn); 906 PJDLOG_ASSERT(fd >= 0); 907 FD_SET(fd, &rfds); 908 fd = proto_descriptor(cfg->hc_listenconn); 909 PJDLOG_ASSERT(fd >= 0); 910 FD_SET(fd, &rfds); 911 maxfd = fd > maxfd ? fd : maxfd; 912 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 913 if (res->hr_event == NULL) 914 continue; 915 PJDLOG_ASSERT(res->hr_conn != NULL); 916 fd = proto_descriptor(res->hr_event); 917 PJDLOG_ASSERT(fd >= 0); 918 FD_SET(fd, &rfds); 919 maxfd = fd > maxfd ? fd : maxfd; 920 if (res->hr_role == HAST_ROLE_PRIMARY) { 921 /* Only primary workers asks for connections. */ 922 fd = proto_descriptor(res->hr_conn); 923 PJDLOG_ASSERT(fd >= 0); 924 FD_SET(fd, &rfds); 925 maxfd = fd > maxfd ? fd : maxfd; 926 } 927 } 928 929 PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE); 930 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 931 if (ret == 0) 932 hook_check(); 933 else if (ret == -1) { 934 if (errno == EINTR) 935 continue; 936 KEEP_ERRNO((void)pidfile_remove(pfh)); 937 pjdlog_exit(EX_OSERR, "select() failed"); 938 } 939 940 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 941 control_handle(cfg); 942 if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds)) 943 listen_accept(); 944 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 945 if (res->hr_event == NULL) 946 continue; 947 PJDLOG_ASSERT(res->hr_conn != NULL); 948 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 949 if (event_recv(res) == 0) 950 continue; 951 /* The worker process exited? */ 952 proto_close(res->hr_event); 953 res->hr_event = NULL; 954 proto_close(res->hr_conn); 955 res->hr_conn = NULL; 956 continue; 957 } 958 if (res->hr_role == HAST_ROLE_PRIMARY && 959 FD_ISSET(proto_descriptor(res->hr_conn), &rfds)) { 960 connection_migrate(res); 961 } 962 } 963 } 964} 965 966static void 967dummy_sighandler(int sig __unused) 968{ 969 /* Nothing to do. */ 970} 971 972int 973main(int argc, char *argv[]) 974{ 975 const char *pidfile; 976 pid_t otherpid; 977 bool foreground; 978 int debuglevel; 979 sigset_t mask; 980 981 foreground = false; 982 debuglevel = 0; 983 pidfile = HASTD_PIDFILE; 984 985 for (;;) { 986 int ch; 987 988 ch = getopt(argc, argv, "c:dFhP:"); 989 if (ch == -1) 990 break; 991 switch (ch) { 992 case 'c': 993 cfgpath = optarg; 994 break; 995 case 'd': 996 debuglevel++; 997 break; 998 case 'F': 999 foreground = true; 1000 break; 1001 case 'P': 1002 pidfile = optarg; 1003 break; 1004 case 'h': 1005 default: 1006 usage(); 1007 } 1008 } 1009 argc -= optind; 1010 argv += optind; 1011 1012 pjdlog_init(PJDLOG_MODE_STD); 1013 pjdlog_debug_set(debuglevel); 1014 1015 g_gate_load(); 1016 1017 pfh = pidfile_open(pidfile, 0600, &otherpid); 1018 if (pfh == NULL) { 1019 if (errno == EEXIST) { 1020 pjdlog_exitx(EX_TEMPFAIL, 1021 "Another hastd is already running, pid: %jd.", 1022 (intmax_t)otherpid); 1023 } 1024 /* If we cannot create pidfile from other reasons, only warn. */ 1025 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile"); 1026 } 1027 1028 cfg = yy_config_parse(cfgpath, true); 1029 PJDLOG_ASSERT(cfg != NULL); 1030 1031 /* 1032 * Restore default actions for interesting signals in case parent 1033 * process (like init(8)) decided to ignore some of them (like SIGHUP). 1034 */ 1035 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 1036 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 1037 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 1038 /* 1039 * Because SIGCHLD is ignored by default, setup dummy handler for it, 1040 * so we can mask it. 1041 */ 1042 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 1043 1044 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1045 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1046 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1047 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1048 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1049 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 1050 1051 /* Listen on control address. */ 1052 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) { 1053 KEEP_ERRNO((void)pidfile_remove(pfh)); 1054 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 1055 cfg->hc_controladdr); 1056 } 1057 /* Listen for remote connections. */ 1058 if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) { 1059 KEEP_ERRNO((void)pidfile_remove(pfh)); 1060 pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 1061 cfg->hc_listenaddr); 1062 } 1063 1064 if (!foreground) { 1065 if (daemon(0, 0) < 0) { 1066 KEEP_ERRNO((void)pidfile_remove(pfh)); 1067 pjdlog_exit(EX_OSERR, "Unable to daemonize"); 1068 } 1069 1070 /* Start logging to syslog. */ 1071 pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 1072 1073 /* Write PID to a file. */ 1074 if (pidfile_write(pfh) < 0) { 1075 pjdlog_errno(LOG_WARNING, 1076 "Unable to write PID to a file"); 1077 } 1078 } 1079 1080 hook_init(); 1081 1082 main_loop(); 1083 1084 exit(0); 1085} 1086