sbin/hastd/hastd.c

   1 /*-
   2  * Copyright (c) 2009-2010 The FreeBSD Foundation
   3  * Copyright (c) 2010 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("$FreeBSD$");
  33
  34 #include <sys/param.h>
  35 #include <sys/linker.h>
  36 #include <sys/module.h>
  37 #include <sys/wait.h>
  38
  39 #include <assert.h>
  40 #include <err.h>
  41 #include <errno.h>
  42 #include <libutil.h>
  43 #include <signal.h>
  44 #include <stdbool.h>
  45 #include <stdio.h>
  46 #include <stdlib.h>
  47 #include <string.h>
  48 #include <sysexits.h>
  49 #include <unistd.h>
  50
  51 #include <activemap.h>
  52 #include <pjdlog.h>
  53
  54 #include "control.h"
  55 #include "event.h"
  56 #include "hast.h"
  57 #include "hast_proto.h"
  58 #include "hastd.h"
  59 #include "hooks.h"
  60 #include "subr.h"
  61
  62 /* Path to configuration file. */
  63 const char *cfgpath = HAST_CONFIG;
  64 /* Hastd configuration. */
  65 static struct hastd_config *cfg;
  66 /* Was SIGINT or SIGTERM signal received? */
  67 bool sigexit_received = false;
  68 /* PID file handle. */
  69 struct pidfh *pfh;
  70
  71 /* How often check for hooks running for too long. */
  72 #define REPORT_INTERVAL 10
  73
  74 static void
  75 usage(void)
  76 {
  77
  78         errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]");
  79 }
  80
  81 static void
  82 g_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
  96 static void
  97 child_exit_log(unsigned int pid, int status)
  98 {
  99
 100         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 101                 pjdlog_debug(1, "Worker process exited gracefully (pid=%u).",
 102                     pid);
 103         } else if (WIFSIGNALED(status)) {
 104                 pjdlog_error("Worker process killed (pid=%u, signal=%d).",
 105                     pid, WTERMSIG(status));
 106         } else {
 107                 pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).",
 108                     pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1);
 109         }
 110 }
 111
 112 static void
 113 child_exit(void)
 114 {
 115         struct hast_resource *res;
 116         int status;
 117         pid_t pid;
 118
 119         while ((pid = wait3(&status, WNOHANG, NULL)) > 0) {
 120                 /* Find resource related to the process that just exited. */
 121                 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
 122                         if (pid == res->hr_workerpid)
 123                                 break;
 124                 }
 125                 if (res == NULL) {
 126                         /*
 127                          * This can happen when new connection arrives and we
 128                          * cancel child responsible for the old one or if this
 129                          * was hook which we executed.
 130                          */
 131                         hook_check_one(pid, status);
 132                         continue;
 133                 }
 134                 pjdlog_prefix_set("[%s] (%s) ", res->hr_name,
 135                     role2str(res->hr_role));
 136                 child_exit_log(pid, status);
 137                 child_cleanup(res);
 138                 if (res->hr_role == HAST_ROLE_PRIMARY) {
 139                         /*
 140                          * Restart child process if it was killed by signal
 141                          * or exited because of temporary problem.
 142                          */
 143                         if (WIFSIGNALED(status) ||
 144                             (WIFEXITED(status) &&
 145                              WEXITSTATUS(status) == EX_TEMPFAIL)) {
 146                                 sleep(1);
 147                                 pjdlog_info("Restarting worker process.");
 148                                 hastd_primary(res);
 149                         } else {
 150                                 res->hr_role = HAST_ROLE_INIT;
 151                                 pjdlog_info("Changing resource role back to %s.",
 152                                     role2str(res->hr_role));
 153                         }
 154                 }
 155                 pjdlog_prefix_set("%s", "");
 156         }
 157 }
 158
 159 static bool
 160 resource_needs_restart(const struct hast_resource *res0,
 161     const struct hast_resource *res1)
 162 {
 163
 164         assert(strcmp(res0->hr_name, res1->hr_name) == 0);
 165
 166         if (strcmp(res0->hr_provname, res1->hr_provname) != 0)
 167                 return (true);
 168         if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0)
 169                 return (true);
 170         if (res0->hr_role == HAST_ROLE_INIT ||
 171             res0->hr_role == HAST_ROLE_SECONDARY) {
 172                 if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
 173                         return (true);
 174                 if (res0->hr_replication != res1->hr_replication)
 175                         return (true);
 176                 if (res0->hr_timeout != res1->hr_timeout)
 177                         return (true);
 178                 if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
 179                         return (true);
 180         }
 181         return (false);
 182 }
 183
 184 static bool
 185 resource_needs_reload(const struct hast_resource *res0,
 186     const struct hast_resource *res1)
 187 {
 188
 189         assert(strcmp(res0->hr_name, res1->hr_name) == 0);
 190         assert(strcmp(res0->hr_provname, res1->hr_provname) == 0);
 191         assert(strcmp(res0->hr_localpath, res1->hr_localpath) == 0);
 192
 193         if (res0->hr_role != HAST_ROLE_PRIMARY)
 194                 return (false);
 195
 196         if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0)
 197                 return (true);
 198         if (res0->hr_replication != res1->hr_replication)
 199                 return (true);
 200         if (res0->hr_timeout != res1->hr_timeout)
 201                 return (true);
 202         if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
 203                 return (true);
 204         return (false);
 205 }
 206
 207 static void
 208 hastd_reload(void)
 209 {
 210         struct hastd_config *newcfg;
 211         struct hast_resource *nres, *cres, *tres;
 212         uint8_t role;
 213
 214         pjdlog_info("Reloading configuration...");
 215
 216         newcfg = yy_config_parse(cfgpath, false);
 217         if (newcfg == NULL)
 218                 goto failed;
 219
 220         /*
 221          * Check if control address has changed.
 222          */
 223         if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) {
 224                 if (proto_server(newcfg->hc_controladdr,
 225                     &newcfg->hc_controlconn) < 0) {
 226                         pjdlog_errno(LOG_ERR,
 227                             "Unable to listen on control address %s",
 228                             newcfg->hc_controladdr);
 229                         goto failed;
 230                 }
 231         }
 232         /*
 233          * Check if listen address has changed.
 234          */
 235         if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) {
 236                 if (proto_server(newcfg->hc_listenaddr,
 237                     &newcfg->hc_listenconn) < 0) {
 238                         pjdlog_errno(LOG_ERR, "Unable to listen on address %s",
 239                             newcfg->hc_listenaddr);
 240                         goto failed;
 241                 }
 242         }
 243         /*
 244          * Only when both control and listen sockets are successfully
 245          * initialized switch them to new configuration.
 246          */
 247         if (newcfg->hc_controlconn != NULL) {
 248                 pjdlog_info("Control socket changed from %s to %s.",
 249                     cfg->hc_controladdr, newcfg->hc_controladdr);
 250                 proto_close(cfg->hc_controlconn);
 251                 cfg->hc_controlconn = newcfg->hc_controlconn;
 252                 newcfg->hc_controlconn = NULL;
 253                 strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr,
 254                     sizeof(cfg->hc_controladdr));
 255         }
 256         if (newcfg->hc_listenconn != NULL) {
 257                 pjdlog_info("Listen socket changed from %s to %s.",
 258                     cfg->hc_listenaddr, newcfg->hc_listenaddr);
 259                 proto_close(cfg->hc_listenconn);
 260                 cfg->hc_listenconn = newcfg->hc_listenconn;
 261                 newcfg->hc_listenconn = NULL;
 262                 strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr,
 263                     sizeof(cfg->hc_listenaddr));
 264         }
 265
 266         /*
 267          * Stop and remove resources that were removed from the configuration.
 268          */
 269         TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) {
 270                 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) {
 271                         if (strcmp(cres->hr_name, nres->hr_name) == 0)
 272                                 break;
 273                 }
 274                 if (nres == NULL) {
 275                         control_set_role(cres, HAST_ROLE_INIT);
 276                         TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
 277                         pjdlog_info("Resource %s removed.", cres->hr_name);
 278                         free(cres);
 279                 }
 280         }
 281         /*
 282          * Move new resources to the current configuration.
 283          */
 284         TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
 285                 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
 286                         if (strcmp(cres->hr_name, nres->hr_name) == 0)
 287                                 break;
 288                 }
 289                 if (cres == NULL) {
 290                         TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
 291                         TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
 292                         pjdlog_info("Resource %s added.", nres->hr_name);
 293                 }
 294         }
 295         /*
 296          * Deal with modified resources.
 297          * Depending on what has changed exactly we might want to perform
 298          * different actions.
 299          *
 300          * We do full resource restart in the following situations:
 301          * Resource role is INIT or SECONDARY.
 302          * Resource role is PRIMARY and path to local component or provider
 303          * name has changed.
 304          * In case of PRIMARY, the worker process will be killed and restarted,
 305          * which also means removing /dev/hast/<name> provider and
 306          * recreating it.
 307          *
 308          * We do just reload (send SIGHUP to worker process) if we act as
 309          * PRIMARY, but only remote address, replication mode and timeout
 310          * has changed. For those, there is no need to restart worker process.
 311          * If PRIMARY receives SIGHUP, it will reconnect if remote address or
 312          * replication mode has changed or simply set new timeout if only
 313          * timeout has changed.
 314          */
 315         TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) {
 316                 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) {
 317                         if (strcmp(cres->hr_name, nres->hr_name) == 0)
 318                                 break;
 319                 }
 320                 assert(cres != NULL);
 321                 if (resource_needs_restart(cres, nres)) {
 322                         pjdlog_info("Resource %s configuration was modified, restarting it.",
 323                             cres->hr_name);
 324                         role = cres->hr_role;
 325                         control_set_role(cres, HAST_ROLE_INIT);
 326                         TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next);
 327                         free(cres);
 328                         TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next);
 329                         TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next);
 330                         control_set_role(nres, role);
 331                 } else if (resource_needs_reload(cres, nres)) {
 332                         pjdlog_info("Resource %s configuration was modified, reloading it.",
 333                             cres->hr_name);
 334                         strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr,
 335                             sizeof(cres->hr_remoteaddr));
 336                         cres->hr_replication = nres->hr_replication;
 337                         cres->hr_timeout = nres->hr_timeout;
 338                         if (cres->hr_workerpid != 0) {
 339                                 if (kill(cres->hr_workerpid, SIGHUP) < 0) {
 340                                         pjdlog_errno(LOG_WARNING,
 341                                             "Unable to send SIGHUP to worker process %u",
 342                                             (unsigned int)cres->hr_workerpid);
 343                                 }
 344                         }
 345                 }
 346         }
 347
 348         yy_config_free(newcfg);
 349         pjdlog_info("Configuration reloaded successfully.");
 350         return;
 351 failed:
 352         if (newcfg != NULL) {
 353                 if (newcfg->hc_controlconn != NULL)
 354                         proto_close(newcfg->hc_controlconn);
 355                 if (newcfg->hc_listenconn != NULL)
 356                         proto_close(newcfg->hc_listenconn);
 357                 yy_config_free(newcfg);
 358         }
 359         pjdlog_warning("Configuration not reloaded.");
 360 }
 361
 362 static void
 363 terminate_workers(void)
 364 {
 365         struct hast_resource *res;
 366
 367         pjdlog_info("Termination signal received, exiting.");
 368         TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
 369                 if (res->hr_workerpid == 0)
 370                         continue;
 371                 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).",
 372                     res->hr_name, role2str(res->hr_role), res->hr_workerpid);
 373                 if (kill(res->hr_workerpid, SIGTERM) == 0)
 374                         continue;
 375                 pjdlog_errno(LOG_WARNING,
 376                     "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).",
 377                     res->hr_name, role2str(res->hr_role), res->hr_workerpid);
 378         }
 379 }
 380
 381 static void
 382 listen_accept(void)
 383 {
 384         struct hast_resource *res;
 385         struct proto_conn *conn;
 386         struct nv *nvin, *nvout, *nverr;
 387         const char *resname;
 388         const unsigned char *token;
 389         char laddr[256], raddr[256];
 390         size_t size;
 391         pid_t pid;
 392         int status;
 393
 394         proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr));
 395         pjdlog_debug(1, "Accepting connection to %s.", laddr);
 396
 397         if (proto_accept(cfg->hc_listenconn, &conn) < 0) {
 398                 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr);
 399                 return;
 400         }
 401
 402         proto_local_address(conn, laddr, sizeof(laddr));
 403         proto_remote_address(conn, raddr, sizeof(raddr));
 404         pjdlog_info("Connection from %s to %s.", raddr, laddr);
 405
 406         /* Error in setting timeout is not critical, but why should it fail? */
 407         if (proto_timeout(conn, HAST_TIMEOUT) < 0)
 408                 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
 409
 410         nvin = nvout = nverr = NULL;
 411
 412         /*
 413          * Before receiving any data see if remote host have access to any
 414          * resource.
 415          */
 416         TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
 417                 if (proto_address_match(conn, res->hr_remoteaddr))
 418                         break;
 419         }
 420         if (res == NULL) {
 421                 pjdlog_error("Client %s isn't known.", raddr);
 422                 goto close;
 423         }
 424         /* Ok, remote host can access at least one resource. */
 425
 426         if (hast_proto_recv_hdr(conn, &nvin) < 0) {
 427                 pjdlog_errno(LOG_ERR, "Unable to receive header from %s",
 428                     raddr);
 429                 goto close;
 430         }
 431
 432         resname = nv_get_string(nvin, "resource");
 433         if (resname == NULL) {
 434                 pjdlog_error("No 'resource' field in the header received from %s.",
 435                     raddr);
 436                 goto close;
 437         }
 438         pjdlog_debug(2, "%s: resource=%s", raddr, resname);
 439         token = nv_get_uint8_array(nvin, &size, "token");
 440         /*
 441          * NULL token means that this is first conection.
 442          */
 443         if (token != NULL && size != sizeof(res->hr_token)) {
 444                 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).",
 445                     raddr, sizeof(res->hr_token), size);
 446                 goto close;
 447         }
 448
 449         /*
 450          * From now on we want to send errors to the remote node.
 451          */
 452         nverr = nv_alloc();
 453
 454         /* Find resource related to this connection. */
 455         TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
 456                 if (strcmp(resname, res->hr_name) == 0)
 457                         break;
 458         }
 459         /* Have we found the resource? */
 460         if (res == NULL) {
 461                 pjdlog_error("No resource '%s' as requested by %s.",
 462                     resname, raddr);
 463                 nv_add_stringf(nverr, "errmsg", "Resource not configured.");
 464                 goto fail;
 465         }
 466
 467         /* Now that we know resource name setup log prefix. */
 468         pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
 469
 470         /* Does the remote host have access to this resource? */
 471         if (!proto_address_match(conn, res->hr_remoteaddr)) {
 472                 pjdlog_error("Client %s has no access to the resource.", raddr);
 473                 nv_add_stringf(nverr, "errmsg", "No access to the resource.");
 474                 goto fail;
 475         }
 476         /* Is the resource marked as secondary? */
 477         if (res->hr_role != HAST_ROLE_SECONDARY) {
 478                 pjdlog_error("We act as %s for the resource and not as %s as requested by %s.",
 479                     role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY),
 480                     raddr);
 481                 nv_add_stringf(nverr, "errmsg",
 482                     "Remote node acts as %s for the resource and not as %s.",
 483                     role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY));
 484                 goto fail;
 485         }
 486         /* Does token (if exists) match? */
 487         if (token != NULL && memcmp(token, res->hr_token,
 488             sizeof(res->hr_token)) != 0) {
 489                 pjdlog_error("Token received from %s doesn't match.", raddr);
 490                 nv_add_stringf(nverr, "errmsg", "Token doesn't match.");
 491                 goto fail;
 492         }
 493         /*
 494          * If there is no token, but we have half-open connection
 495          * (only remotein) or full connection (worker process is running)
 496          * we have to cancel those and accept the new connection.
 497          */
 498         if (token == NULL) {
 499                 assert(res->hr_remoteout == NULL);
 500                 pjdlog_debug(1, "Initial connection from %s.", raddr);
 501                 if (res->hr_workerpid != 0) {
 502                         assert(res->hr_remotein == NULL);
 503                         pjdlog_debug(1,
 504                             "Worker process exists (pid=%u), stopping it.",
 505                             (unsigned int)res->hr_workerpid);
 506                         /* Stop child process. */
 507                         if (kill(res->hr_workerpid, SIGINT) < 0) {
 508                                 pjdlog_errno(LOG_ERR,
 509                                     "Unable to stop worker process (pid=%u)",
 510                                     (unsigned int)res->hr_workerpid);
 511                                 /*
 512                                  * Other than logging the problem we
 513                                  * ignore it - nothing smart to do.
 514                                  */
 515                         }
 516                         /* Wait for it to exit. */
 517                         else if ((pid = waitpid(res->hr_workerpid,
 518                             &status, 0)) != res->hr_workerpid) {
 519                                 /* We can only log the problem. */
 520                                 pjdlog_errno(LOG_ERR,
 521                                     "Waiting for worker process (pid=%u) failed",
 522                                     (unsigned int)res->hr_workerpid);
 523                         } else {
 524                                 child_exit_log(res->hr_workerpid, status);
 525                         }
 526                         child_cleanup(res);
 527                 } else if (res->hr_remotein != NULL) {
 528                         char oaddr[256];
 529
 530                         proto_remote_address(conn, oaddr, sizeof(oaddr));
 531                         pjdlog_debug(1,
 532                             "Canceling half-open connection from %s on connection from %s.",
 533                             oaddr, raddr);
 534                         proto_close(res->hr_remotein);
 535                         res->hr_remotein = NULL;
 536                 }
 537         }
 538
 539         /*
 540          * Checks and cleanups are done.
 541          */
 542
 543         if (token == NULL) {
 544                 arc4random_buf(res->hr_token, sizeof(res->hr_token));
 545                 nvout = nv_alloc();
 546                 nv_add_uint8_array(nvout, res->hr_token,
 547                     sizeof(res->hr_token), "token");
 548                 if (nv_error(nvout) != 0) {
 549                         pjdlog_common(LOG_ERR, 0, nv_error(nvout),
 550                             "Unable to prepare return header for %s", raddr);
 551                         nv_add_stringf(nverr, "errmsg",
 552                             "Remote node was unable to prepare return header: %s.",
 553                             strerror(nv_error(nvout)));
 554                         goto fail;
 555                 }
 556                 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) {
 557                         int error = errno;
 558
 559                         pjdlog_errno(LOG_ERR, "Unable to send response to %s",
 560                             raddr);
 561                         nv_add_stringf(nverr, "errmsg",
 562                             "Remote node was unable to send response: %s.",
 563                             strerror(error));
 564                         goto fail;
 565                 }
 566                 res->hr_remotein = conn;
 567                 pjdlog_debug(1, "Incoming connection from %s configured.",
 568                     raddr);
 569         } else {
 570                 res->hr_remoteout = conn;
 571                 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr);
 572                 hastd_secondary(res, nvin);
 573         }
 574         nv_free(nvin);
 575         nv_free(nvout);
 576         nv_free(nverr);
 577         pjdlog_prefix_set("%s", "");
 578         return;
 579 fail:
 580         if (nv_error(nverr) != 0) {
 581                 pjdlog_common(LOG_ERR, 0, nv_error(nverr),
 582                     "Unable to prepare error header for %s", raddr);
 583                 goto close;
 584         }
 585         if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) {
 586                 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr);
 587                 goto close;
 588         }
 589 close:
 590         if (nvin != NULL)
 591                 nv_free(nvin);
 592         if (nvout != NULL)
 593                 nv_free(nvout);
 594         if (nverr != NULL)
 595                 nv_free(nverr);
 596         proto_close(conn);
 597         pjdlog_prefix_set("%s", "");
 598 }
 599
 600 static void
 601 main_loop(void)
 602 {
 603         struct hast_resource *res;
 604         struct timeval seltimeout;
 605         struct timespec sigtimeout;
 606         int fd, maxfd, ret, signo;
 607         sigset_t mask;
 608         fd_set rfds;
 609
 610         seltimeout.tv_sec = REPORT_INTERVAL;
 611         seltimeout.tv_usec = 0;
 612         sigtimeout.tv_sec = 0;
 613         sigtimeout.tv_nsec = 0;
 614
 615         PJDLOG_VERIFY(sigemptyset(&mask) == 0);
 616         PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
 617         PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
 618         PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
 619         PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
 620
 621         for (;;) {
 622                 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) {
 623                         switch (signo) {
 624                         case SIGINT:
 625                         case SIGTERM:
 626                                 sigexit_received = true;
 627                                 terminate_workers();
 628                                 exit(EX_OK);
 629                                 break;
 630                         case SIGCHLD:
 631                                 child_exit();
 632                                 break;
 633                         case SIGHUP:
 634                                 hastd_reload();
 635                                 break;
 636                         default:
 637                                 assert(!"invalid condition");
 638                         }
 639                 }
 640
 641                 /* Setup descriptors for select(2). */
 642                 FD_ZERO(&rfds);
 643                 maxfd = fd = proto_descriptor(cfg->hc_controlconn);
 644                 assert(fd >= 0);
 645                 FD_SET(fd, &rfds);
 646                 fd = proto_descriptor(cfg->hc_listenconn);
 647                 assert(fd >= 0);
 648                 FD_SET(fd, &rfds);
 649                 maxfd = fd > maxfd ? fd : maxfd;
 650                 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
 651                         if (res->hr_event == NULL)
 652                                 continue;
 653                         fd = proto_descriptor(res->hr_event);
 654                         assert(fd >= 0);
 655                         FD_SET(fd, &rfds);
 656                         maxfd = fd > maxfd ? fd : maxfd;
 657                 }
 658
 659                 assert(maxfd + 1 <= (int)FD_SETSIZE);
 660                 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout);
 661                 if (ret == 0)
 662                         hook_check(false);
 663                 else if (ret == -1) {
 664                         if (errno == EINTR)
 665                                 continue;
 666                         KEEP_ERRNO((void)pidfile_remove(pfh));
 667                         pjdlog_exit(EX_OSERR, "select() failed");
 668                 }
 669
 670                 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds))
 671                         control_handle(cfg);
 672                 if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds))
 673                         listen_accept();
 674                 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
 675                         if (res->hr_event == NULL)
 676                                 continue;
 677                         if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) {
 678                                 if (event_recv(res) == 0)
 679                                         continue;
 680                                 /* The worker process exited? */
 681                                 proto_close(res->hr_event);
 682                                 res->hr_event = NULL;
 683                         }
 684                 }
 685         }
 686 }
 687
 688 int
 689 main(int argc, char *argv[])
 690 {
 691         const char *pidfile;
 692         pid_t otherpid;
 693         bool foreground;
 694         int debuglevel;
 695         sigset_t mask;
 696
 697         g_gate_load();
 698
 699         foreground = false;
 700         debuglevel = 0;
 701         pidfile = HASTD_PIDFILE;
 702
 703         for (;;) {
 704                 int ch;
 705
 706                 ch = getopt(argc, argv, "c:dFhP:");
 707                 if (ch == -1)
 708                         break;
 709                 switch (ch) {
 710                 case 'c':
 711                         cfgpath = optarg;
 712                         break;
 713                 case 'd':
 714                         debuglevel++;
 715                         break;
 716                 case 'F':
 717                         foreground = true;
 718                         break;
 719                 case 'P':
 720                         pidfile = optarg;
 721                         break;
 722                 case 'h':
 723                 default:
 724                         usage();
 725                 }
 726         }
 727         argc -= optind;
 728         argv += optind;
 729
 730         pjdlog_debug_set(debuglevel);
 731
 732         pfh = pidfile_open(pidfile, 0600, &otherpid);
 733         if (pfh == NULL) {
 734                 if (errno == EEXIST) {
 735                         pjdlog_exitx(EX_TEMPFAIL,
 736                             "Another hastd is already running, pid: %jd.",
 737                             (intmax_t)otherpid);
 738                 }
 739                 /* If we cannot create pidfile from other reasons, only warn. */
 740                 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile");
 741         }
 742
 743         cfg = yy_config_parse(cfgpath, true);
 744         assert(cfg != NULL);
 745
 746         PJDLOG_VERIFY(sigemptyset(&mask) == 0);
 747         PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0);
 748         PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
 749         PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
 750         PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0);
 751         PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 752
 753         /* Listen on control address. */
 754         if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) {
 755                 KEEP_ERRNO((void)pidfile_remove(pfh));
 756                 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s",
 757                     cfg->hc_controladdr);
 758         }
 759         /* Listen for remote connections. */
 760         if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) {
 761                 KEEP_ERRNO((void)pidfile_remove(pfh));
 762                 pjdlog_exit(EX_OSERR, "Unable to listen on address %s",
 763                     cfg->hc_listenaddr);
 764         }
 765
 766         if (!foreground) {
 767                 if (daemon(0, 0) < 0) {
 768                         KEEP_ERRNO((void)pidfile_remove(pfh));
 769                         pjdlog_exit(EX_OSERR, "Unable to daemonize");
 770                 }
 771
 772                 /* Start logging to syslog. */
 773                 pjdlog_mode_set(PJDLOG_MODE_SYSLOG);
 774
 775                 /* Write PID to a file. */
 776                 if (pidfile_write(pfh) < 0) {
 777                         pjdlog_errno(LOG_WARNING,
 778                             "Unable to write PID to a file");
 779                 }
 780         }
 781
 782         hook_init();
 783
 784         main_loop();
 785
 786         exit(0);
 787 }