sbin/hastd/hastd.c

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