hastd.c revision 219354
1204076Spjd/*- 2204076Spjd * Copyright (c) 2009-2010 The FreeBSD Foundation 3219351Spjd * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net> 4204076Spjd * All rights reserved. 5204076Spjd * 6204076Spjd * This software was developed by Pawel Jakub Dawidek under sponsorship from 7204076Spjd * the FreeBSD Foundation. 8204076Spjd * 9204076Spjd * Redistribution and use in source and binary forms, with or without 10204076Spjd * modification, are permitted provided that the following conditions 11204076Spjd * are met: 12204076Spjd * 1. Redistributions of source code must retain the above copyright 13204076Spjd * notice, this list of conditions and the following disclaimer. 14204076Spjd * 2. Redistributions in binary form must reproduce the above copyright 15204076Spjd * notice, this list of conditions and the following disclaimer in the 16204076Spjd * documentation and/or other materials provided with the distribution. 17204076Spjd * 18204076Spjd * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 19204076Spjd * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20204076Spjd * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21204076Spjd * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 22204076Spjd * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23204076Spjd * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24204076Spjd * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25204076Spjd * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26204076Spjd * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27204076Spjd * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28204076Spjd * SUCH DAMAGE. 29204076Spjd */ 30204076Spjd 31204076Spjd#include <sys/cdefs.h> 32204076Spjd__FBSDID("$FreeBSD: head/sbin/hastd/hastd.c 219354 2011-03-06 23:09:33Z pjd $"); 33204076Spjd 34204076Spjd#include <sys/param.h> 35204076Spjd#include <sys/linker.h> 36204076Spjd#include <sys/module.h> 37218044Spjd#include <sys/stat.h> 38204076Spjd#include <sys/wait.h> 39204076Spjd 40204076Spjd#include <err.h> 41204076Spjd#include <errno.h> 42204076Spjd#include <libutil.h> 43204076Spjd#include <signal.h> 44204076Spjd#include <stdbool.h> 45204076Spjd#include <stdio.h> 46204076Spjd#include <stdlib.h> 47204076Spjd#include <string.h> 48204076Spjd#include <sysexits.h> 49204076Spjd#include <unistd.h> 50204076Spjd 51204076Spjd#include <activemap.h> 52204076Spjd#include <pjdlog.h> 53204076Spjd 54204076Spjd#include "control.h" 55212038Spjd#include "event.h" 56204076Spjd#include "hast.h" 57204076Spjd#include "hast_proto.h" 58204076Spjd#include "hastd.h" 59211977Spjd#include "hooks.h" 60204076Spjd#include "subr.h" 61204076Spjd 62204076Spjd/* Path to configuration file. */ 63210886Spjdconst char *cfgpath = HAST_CONFIG; 64204076Spjd/* Hastd configuration. */ 65204076Spjdstatic struct hastd_config *cfg; 66204076Spjd/* Was SIGINT or SIGTERM signal received? */ 67204076Spjdbool sigexit_received = false; 68204076Spjd/* PID file handle. */ 69204076Spjdstruct pidfh *pfh; 70204076Spjd 71211977Spjd/* How often check for hooks running for too long. */ 72213430Spjd#define REPORT_INTERVAL 5 73211977Spjd 74204076Spjdstatic void 75204076Spjdusage(void) 76204076Spjd{ 77204076Spjd 78204076Spjd errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]"); 79204076Spjd} 80204076Spjd 81204076Spjdstatic void 82204076Spjdg_gate_load(void) 83204076Spjd{ 84204076Spjd 85204076Spjd if (modfind("g_gate") == -1) { 86204076Spjd /* Not present in kernel, try loading it. */ 87204076Spjd if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) { 88204076Spjd if (errno != EEXIST) { 89204076Spjd pjdlog_exit(EX_OSERR, 90204076Spjd "Unable to load geom_gate module"); 91204076Spjd } 92204076Spjd } 93204076Spjd } 94204076Spjd} 95204076Spjd 96218041Spjdvoid 97218041Spjddescriptors_cleanup(struct hast_resource *res) 98218041Spjd{ 99218041Spjd struct hast_resource *tres; 100218041Spjd 101218041Spjd TAILQ_FOREACH(tres, &cfg->hc_resources, hr_next) { 102218041Spjd if (tres == res) { 103218041Spjd PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY || 104218041Spjd (res->hr_remotein == NULL && 105218041Spjd res->hr_remoteout == NULL)); 106218041Spjd continue; 107218041Spjd } 108218041Spjd if (tres->hr_remotein != NULL) 109218041Spjd proto_close(tres->hr_remotein); 110218041Spjd if (tres->hr_remoteout != NULL) 111218041Spjd proto_close(tres->hr_remoteout); 112218370Spjd if (tres->hr_ctrl != NULL) 113218370Spjd proto_close(tres->hr_ctrl); 114218370Spjd if (tres->hr_event != NULL) 115218370Spjd proto_close(tres->hr_event); 116218370Spjd if (tres->hr_conn != NULL) 117218370Spjd proto_close(tres->hr_conn); 118218041Spjd } 119218041Spjd if (cfg->hc_controlin != NULL) 120218041Spjd proto_close(cfg->hc_controlin); 121218041Spjd proto_close(cfg->hc_controlconn); 122218041Spjd proto_close(cfg->hc_listenconn); 123218041Spjd (void)pidfile_close(pfh); 124218041Spjd hook_fini(); 125218041Spjd pjdlog_fini(); 126218041Spjd} 127218041Spjd 128218044Spjdstatic const char * 129218044Spjddtype2str(mode_t mode) 130218044Spjd{ 131218044Spjd 132218044Spjd if (S_ISBLK(mode)) 133218044Spjd return ("block device"); 134218044Spjd else if (S_ISCHR(mode)) 135218044Spjd return ("character device"); 136218044Spjd else if (S_ISDIR(mode)) 137218044Spjd return ("directory"); 138218044Spjd else if (S_ISFIFO(mode)) 139218044Spjd return ("pipe or FIFO"); 140218044Spjd else if (S_ISLNK(mode)) 141218044Spjd return ("symbolic link"); 142218044Spjd else if (S_ISREG(mode)) 143218044Spjd return ("regular file"); 144218044Spjd else if (S_ISSOCK(mode)) 145218044Spjd return ("socket"); 146218044Spjd else if (S_ISWHT(mode)) 147218044Spjd return ("whiteout"); 148218044Spjd else 149218044Spjd return ("unknown"); 150218044Spjd} 151218044Spjd 152218044Spjdvoid 153218044Spjddescriptors_assert(const struct hast_resource *res, int pjdlogmode) 154218044Spjd{ 155218044Spjd char msg[256]; 156218044Spjd struct stat sb; 157218044Spjd long maxfd; 158218044Spjd bool isopen; 159218044Spjd mode_t mode; 160218044Spjd int fd; 161218044Spjd 162218044Spjd /* 163218044Spjd * At this point descriptor to syslog socket is closed, so if we want 164218044Spjd * to log assertion message, we have to first store it in 'msg' local 165218044Spjd * buffer and then open syslog socket and log it. 166218044Spjd */ 167218044Spjd msg[0] = '\0'; 168218044Spjd 169218044Spjd maxfd = sysconf(_SC_OPEN_MAX); 170218044Spjd if (maxfd < 0) { 171218373Spjd pjdlog_init(pjdlogmode); 172218373Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 173218373Spjd role2str(res->hr_role)); 174218044Spjd pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed"); 175218373Spjd pjdlog_fini(); 176218044Spjd maxfd = 16384; 177218044Spjd } 178218044Spjd for (fd = 0; fd <= maxfd; fd++) { 179218044Spjd if (fstat(fd, &sb) == 0) { 180218044Spjd isopen = true; 181218044Spjd mode = sb.st_mode; 182218044Spjd } else if (errno == EBADF) { 183218044Spjd isopen = false; 184218044Spjd mode = 0; 185218044Spjd } else { 186218375Spjd (void)snprintf(msg, sizeof(msg), 187218044Spjd "Unable to fstat descriptor %d: %s", fd, 188218044Spjd strerror(errno)); 189218374Spjd break; 190218044Spjd } 191218044Spjd if (fd == STDIN_FILENO || fd == STDOUT_FILENO || 192218044Spjd fd == STDERR_FILENO) { 193218044Spjd if (!isopen) { 194218375Spjd (void)snprintf(msg, sizeof(msg), 195218044Spjd "Descriptor %d (%s) is closed, but should be open.", 196218044Spjd fd, (fd == STDIN_FILENO ? "stdin" : 197218044Spjd (fd == STDOUT_FILENO ? "stdout" : "stderr"))); 198218044Spjd break; 199218044Spjd } 200218044Spjd } else if (fd == proto_descriptor(res->hr_event)) { 201218044Spjd if (!isopen) { 202218375Spjd (void)snprintf(msg, sizeof(msg), 203218044Spjd "Descriptor %d (event) is closed, but should be open.", 204218044Spjd fd); 205218044Spjd break; 206218044Spjd } 207218044Spjd if (!S_ISSOCK(mode)) { 208218375Spjd (void)snprintf(msg, sizeof(msg), 209218044Spjd "Descriptor %d (event) is %s, but should be %s.", 210218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 211218044Spjd break; 212218044Spjd } 213218044Spjd } else if (fd == proto_descriptor(res->hr_ctrl)) { 214218044Spjd if (!isopen) { 215218375Spjd (void)snprintf(msg, sizeof(msg), 216218044Spjd "Descriptor %d (ctrl) is closed, but should be open.", 217218044Spjd fd); 218218044Spjd break; 219218044Spjd } 220218044Spjd if (!S_ISSOCK(mode)) { 221218375Spjd (void)snprintf(msg, sizeof(msg), 222218044Spjd "Descriptor %d (ctrl) is %s, but should be %s.", 223218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 224218044Spjd break; 225218044Spjd } 226218218Spjd } else if (fd == proto_descriptor(res->hr_conn)) { 227218218Spjd if (!isopen) { 228218375Spjd (void)snprintf(msg, sizeof(msg), 229218218Spjd "Descriptor %d (conn) is closed, but should be open.", 230218218Spjd fd); 231218218Spjd break; 232218218Spjd } 233218218Spjd if (!S_ISSOCK(mode)) { 234218375Spjd (void)snprintf(msg, sizeof(msg), 235218218Spjd "Descriptor %d (conn) is %s, but should be %s.", 236218218Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 237218218Spjd break; 238218218Spjd } 239218044Spjd } else if (res->hr_role == HAST_ROLE_SECONDARY && 240218044Spjd fd == proto_descriptor(res->hr_remotein)) { 241218044Spjd if (!isopen) { 242218375Spjd (void)snprintf(msg, sizeof(msg), 243218044Spjd "Descriptor %d (remote in) is closed, but should be open.", 244218044Spjd fd); 245218044Spjd break; 246218044Spjd } 247218044Spjd if (!S_ISSOCK(mode)) { 248218375Spjd (void)snprintf(msg, sizeof(msg), 249218044Spjd "Descriptor %d (remote in) is %s, but should be %s.", 250218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 251218044Spjd break; 252218044Spjd } 253218044Spjd } else if (res->hr_role == HAST_ROLE_SECONDARY && 254218044Spjd fd == proto_descriptor(res->hr_remoteout)) { 255218044Spjd if (!isopen) { 256218375Spjd (void)snprintf(msg, sizeof(msg), 257218044Spjd "Descriptor %d (remote out) is closed, but should be open.", 258218044Spjd fd); 259218044Spjd break; 260218044Spjd } 261218044Spjd if (!S_ISSOCK(mode)) { 262218375Spjd (void)snprintf(msg, sizeof(msg), 263218044Spjd "Descriptor %d (remote out) is %s, but should be %s.", 264218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 265218044Spjd break; 266218044Spjd } 267218044Spjd } else { 268218044Spjd if (isopen) { 269218375Spjd (void)snprintf(msg, sizeof(msg), 270218044Spjd "Descriptor %d is open (%s), but should be closed.", 271218044Spjd fd, dtype2str(mode)); 272218044Spjd break; 273218044Spjd } 274218044Spjd } 275218044Spjd } 276218044Spjd if (msg[0] != '\0') { 277218044Spjd pjdlog_init(pjdlogmode); 278218044Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 279218044Spjd role2str(res->hr_role)); 280218044Spjd PJDLOG_ABORT("%s", msg); 281218044Spjd } 282218044Spjd} 283218044Spjd 284204076Spjdstatic void 285207372Spjdchild_exit_log(unsigned int pid, int status) 286207372Spjd{ 287207372Spjd 288207372Spjd if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 289207372Spjd pjdlog_debug(1, "Worker process exited gracefully (pid=%u).", 290207372Spjd pid); 291207372Spjd } else if (WIFSIGNALED(status)) { 292207372Spjd pjdlog_error("Worker process killed (pid=%u, signal=%d).", 293207372Spjd pid, WTERMSIG(status)); 294207372Spjd } else { 295207372Spjd pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).", 296207372Spjd pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1); 297207372Spjd } 298207372Spjd} 299207372Spjd 300207372Spjdstatic void 301204076Spjdchild_exit(void) 302204076Spjd{ 303204076Spjd struct hast_resource *res; 304204076Spjd int status; 305204076Spjd pid_t pid; 306204076Spjd 307204076Spjd while ((pid = wait3(&status, WNOHANG, NULL)) > 0) { 308204076Spjd /* Find resource related to the process that just exited. */ 309204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 310204076Spjd if (pid == res->hr_workerpid) 311204076Spjd break; 312204076Spjd } 313204076Spjd if (res == NULL) { 314204076Spjd /* 315204076Spjd * This can happen when new connection arrives and we 316211977Spjd * cancel child responsible for the old one or if this 317211977Spjd * was hook which we executed. 318204076Spjd */ 319211977Spjd hook_check_one(pid, status); 320204076Spjd continue; 321204076Spjd } 322204076Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 323204076Spjd role2str(res->hr_role)); 324207372Spjd child_exit_log(pid, status); 325213006Spjd child_cleanup(res); 326204076Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 327207372Spjd /* 328207372Spjd * Restart child process if it was killed by signal 329207372Spjd * or exited because of temporary problem. 330207372Spjd */ 331207372Spjd if (WIFSIGNALED(status) || 332207372Spjd (WIFEXITED(status) && 333207372Spjd WEXITSTATUS(status) == EX_TEMPFAIL)) { 334207348Spjd sleep(1); 335207348Spjd pjdlog_info("Restarting worker process."); 336207348Spjd hastd_primary(res); 337207348Spjd } else { 338207348Spjd res->hr_role = HAST_ROLE_INIT; 339207348Spjd pjdlog_info("Changing resource role back to %s.", 340207348Spjd role2str(res->hr_role)); 341207348Spjd } 342204076Spjd } 343204076Spjd pjdlog_prefix_set("%s", ""); 344204076Spjd } 345204076Spjd} 346204076Spjd 347210886Spjdstatic bool 348210886Spjdresource_needs_restart(const struct hast_resource *res0, 349210886Spjd const struct hast_resource *res1) 350210886Spjd{ 351210886Spjd 352218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 353210886Spjd 354210886Spjd if (strcmp(res0->hr_provname, res1->hr_provname) != 0) 355210886Spjd return (true); 356210886Spjd if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0) 357210886Spjd return (true); 358210886Spjd if (res0->hr_role == HAST_ROLE_INIT || 359210886Spjd res0->hr_role == HAST_ROLE_SECONDARY) { 360210886Spjd if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 361210886Spjd return (true); 362210886Spjd if (res0->hr_replication != res1->hr_replication) 363210886Spjd return (true); 364219351Spjd if (res0->hr_checksum != res1->hr_checksum) 365219351Spjd return (true); 366219354Spjd if (res0->hr_compression != res1->hr_compression) 367219354Spjd return (true); 368210886Spjd if (res0->hr_timeout != res1->hr_timeout) 369210886Spjd return (true); 370211886Spjd if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 371211886Spjd return (true); 372210886Spjd } 373210886Spjd return (false); 374210886Spjd} 375210886Spjd 376210886Spjdstatic bool 377210886Spjdresource_needs_reload(const struct hast_resource *res0, 378210886Spjd const struct hast_resource *res1) 379210886Spjd{ 380210886Spjd 381218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 382218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0); 383218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0); 384210886Spjd 385210886Spjd if (res0->hr_role != HAST_ROLE_PRIMARY) 386210886Spjd return (false); 387210886Spjd 388210886Spjd if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 389210886Spjd return (true); 390210886Spjd if (res0->hr_replication != res1->hr_replication) 391210886Spjd return (true); 392219351Spjd if (res0->hr_checksum != res1->hr_checksum) 393219351Spjd return (true); 394219354Spjd if (res0->hr_compression != res1->hr_compression) 395219354Spjd return (true); 396210886Spjd if (res0->hr_timeout != res1->hr_timeout) 397210886Spjd return (true); 398211886Spjd if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 399211886Spjd return (true); 400210886Spjd return (false); 401210886Spjd} 402210886Spjd 403204076Spjdstatic void 404217784Spjdresource_reload(const struct hast_resource *res) 405217784Spjd{ 406217784Spjd struct nv *nvin, *nvout; 407217784Spjd int error; 408217784Spjd 409218138Spjd PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 410217784Spjd 411217784Spjd nvout = nv_alloc(); 412217784Spjd nv_add_uint8(nvout, HASTCTL_RELOAD, "cmd"); 413217784Spjd nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr"); 414217784Spjd nv_add_int32(nvout, (int32_t)res->hr_replication, "replication"); 415219351Spjd nv_add_int32(nvout, (int32_t)res->hr_checksum, "checksum"); 416219354Spjd nv_add_int32(nvout, (int32_t)res->hr_compression, "compression"); 417217784Spjd nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout"); 418217784Spjd nv_add_string(nvout, res->hr_exec, "exec"); 419217784Spjd if (nv_error(nvout) != 0) { 420217784Spjd nv_free(nvout); 421217784Spjd pjdlog_error("Unable to allocate header for reload message."); 422217784Spjd return; 423217784Spjd } 424217784Spjd if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) { 425217784Spjd pjdlog_errno(LOG_ERR, "Unable to send reload message"); 426217784Spjd nv_free(nvout); 427217784Spjd return; 428217784Spjd } 429217784Spjd nv_free(nvout); 430217784Spjd 431217784Spjd /* Receive response. */ 432217784Spjd if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) { 433217784Spjd pjdlog_errno(LOG_ERR, "Unable to receive reload reply"); 434217784Spjd return; 435217784Spjd } 436217784Spjd error = nv_get_int16(nvin, "error"); 437217784Spjd nv_free(nvin); 438217784Spjd if (error != 0) { 439217784Spjd pjdlog_common(LOG_ERR, 0, error, "Reload failed"); 440217784Spjd return; 441217784Spjd } 442217784Spjd} 443217784Spjd 444217784Spjdstatic void 445204076Spjdhastd_reload(void) 446204076Spjd{ 447210886Spjd struct hastd_config *newcfg; 448210886Spjd struct hast_resource *nres, *cres, *tres; 449210886Spjd uint8_t role; 450204076Spjd 451210886Spjd pjdlog_info("Reloading configuration..."); 452210886Spjd 453210886Spjd newcfg = yy_config_parse(cfgpath, false); 454210886Spjd if (newcfg == NULL) 455210886Spjd goto failed; 456210886Spjd 457210886Spjd /* 458210886Spjd * Check if control address has changed. 459210886Spjd */ 460210886Spjd if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 461210886Spjd if (proto_server(newcfg->hc_controladdr, 462210886Spjd &newcfg->hc_controlconn) < 0) { 463210886Spjd pjdlog_errno(LOG_ERR, 464210886Spjd "Unable to listen on control address %s", 465210886Spjd newcfg->hc_controladdr); 466210886Spjd goto failed; 467210886Spjd } 468210886Spjd } 469210886Spjd /* 470210886Spjd * Check if listen address has changed. 471210886Spjd */ 472210886Spjd if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) { 473210886Spjd if (proto_server(newcfg->hc_listenaddr, 474210886Spjd &newcfg->hc_listenconn) < 0) { 475210886Spjd pjdlog_errno(LOG_ERR, "Unable to listen on address %s", 476210886Spjd newcfg->hc_listenaddr); 477210886Spjd goto failed; 478210886Spjd } 479210886Spjd } 480210886Spjd /* 481210886Spjd * Only when both control and listen sockets are successfully 482210886Spjd * initialized switch them to new configuration. 483210886Spjd */ 484210886Spjd if (newcfg->hc_controlconn != NULL) { 485210886Spjd pjdlog_info("Control socket changed from %s to %s.", 486210886Spjd cfg->hc_controladdr, newcfg->hc_controladdr); 487210886Spjd proto_close(cfg->hc_controlconn); 488210886Spjd cfg->hc_controlconn = newcfg->hc_controlconn; 489210886Spjd newcfg->hc_controlconn = NULL; 490210886Spjd strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 491210886Spjd sizeof(cfg->hc_controladdr)); 492210886Spjd } 493210886Spjd if (newcfg->hc_listenconn != NULL) { 494210886Spjd pjdlog_info("Listen socket changed from %s to %s.", 495210886Spjd cfg->hc_listenaddr, newcfg->hc_listenaddr); 496210886Spjd proto_close(cfg->hc_listenconn); 497210886Spjd cfg->hc_listenconn = newcfg->hc_listenconn; 498210886Spjd newcfg->hc_listenconn = NULL; 499210886Spjd strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr, 500210886Spjd sizeof(cfg->hc_listenaddr)); 501210886Spjd } 502210886Spjd 503210886Spjd /* 504210886Spjd * Stop and remove resources that were removed from the configuration. 505210886Spjd */ 506210886Spjd TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 507210886Spjd TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 508210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 509210886Spjd break; 510210886Spjd } 511210886Spjd if (nres == NULL) { 512210886Spjd control_set_role(cres, HAST_ROLE_INIT); 513210886Spjd TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 514210886Spjd pjdlog_info("Resource %s removed.", cres->hr_name); 515210886Spjd free(cres); 516210886Spjd } 517210886Spjd } 518210886Spjd /* 519210886Spjd * Move new resources to the current configuration. 520210886Spjd */ 521210886Spjd TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 522210886Spjd TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 523210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 524210886Spjd break; 525210886Spjd } 526210886Spjd if (cres == NULL) { 527210886Spjd TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 528210886Spjd TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 529210886Spjd pjdlog_info("Resource %s added.", nres->hr_name); 530210886Spjd } 531210886Spjd } 532210886Spjd /* 533210886Spjd * Deal with modified resources. 534210886Spjd * Depending on what has changed exactly we might want to perform 535210886Spjd * different actions. 536210886Spjd * 537210886Spjd * We do full resource restart in the following situations: 538210886Spjd * Resource role is INIT or SECONDARY. 539210886Spjd * Resource role is PRIMARY and path to local component or provider 540210886Spjd * name has changed. 541210886Spjd * In case of PRIMARY, the worker process will be killed and restarted, 542210886Spjd * which also means removing /dev/hast/<name> provider and 543210886Spjd * recreating it. 544210886Spjd * 545210886Spjd * We do just reload (send SIGHUP to worker process) if we act as 546217729Spjd * PRIMARY, but only if remote address, replication mode, timeout or 547217729Spjd * execution path has changed. For those, there is no need to restart 548217729Spjd * worker process. 549210886Spjd * If PRIMARY receives SIGHUP, it will reconnect if remote address or 550210886Spjd * replication mode has changed or simply set new timeout if only 551210886Spjd * timeout has changed. 552210886Spjd */ 553210886Spjd TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 554210886Spjd TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 555210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 556210886Spjd break; 557210886Spjd } 558218138Spjd PJDLOG_ASSERT(cres != NULL); 559210886Spjd if (resource_needs_restart(cres, nres)) { 560210886Spjd pjdlog_info("Resource %s configuration was modified, restarting it.", 561210886Spjd cres->hr_name); 562210886Spjd role = cres->hr_role; 563210886Spjd control_set_role(cres, HAST_ROLE_INIT); 564210886Spjd TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 565210886Spjd free(cres); 566210886Spjd TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 567210886Spjd TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 568210886Spjd control_set_role(nres, role); 569210886Spjd } else if (resource_needs_reload(cres, nres)) { 570210886Spjd pjdlog_info("Resource %s configuration was modified, reloading it.", 571210886Spjd cres->hr_name); 572210886Spjd strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 573210886Spjd sizeof(cres->hr_remoteaddr)); 574210886Spjd cres->hr_replication = nres->hr_replication; 575219351Spjd cres->hr_checksum = nres->hr_checksum; 576219354Spjd cres->hr_compression = nres->hr_compression; 577210886Spjd cres->hr_timeout = nres->hr_timeout; 578217729Spjd strlcpy(cres->hr_exec, nres->hr_exec, 579217729Spjd sizeof(cres->hr_exec)); 580217784Spjd if (cres->hr_workerpid != 0) 581217784Spjd resource_reload(cres); 582210886Spjd } 583210886Spjd } 584210886Spjd 585210886Spjd yy_config_free(newcfg); 586210886Spjd pjdlog_info("Configuration reloaded successfully."); 587210886Spjd return; 588210886Spjdfailed: 589210886Spjd if (newcfg != NULL) { 590210886Spjd if (newcfg->hc_controlconn != NULL) 591210886Spjd proto_close(newcfg->hc_controlconn); 592210886Spjd if (newcfg->hc_listenconn != NULL) 593210886Spjd proto_close(newcfg->hc_listenconn); 594210886Spjd yy_config_free(newcfg); 595210886Spjd } 596210886Spjd pjdlog_warning("Configuration not reloaded."); 597204076Spjd} 598204076Spjd 599204076Spjdstatic void 600211899Spjdterminate_workers(void) 601211899Spjd{ 602211899Spjd struct hast_resource *res; 603211899Spjd 604211899Spjd pjdlog_info("Termination signal received, exiting."); 605211899Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 606211899Spjd if (res->hr_workerpid == 0) 607211899Spjd continue; 608211899Spjd pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 609211899Spjd res->hr_name, role2str(res->hr_role), res->hr_workerpid); 610211899Spjd if (kill(res->hr_workerpid, SIGTERM) == 0) 611211899Spjd continue; 612211899Spjd pjdlog_errno(LOG_WARNING, 613211899Spjd "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 614211899Spjd res->hr_name, role2str(res->hr_role), res->hr_workerpid); 615211899Spjd } 616211899Spjd} 617211899Spjd 618211899Spjdstatic void 619204076Spjdlisten_accept(void) 620204076Spjd{ 621204076Spjd struct hast_resource *res; 622204076Spjd struct proto_conn *conn; 623204076Spjd struct nv *nvin, *nvout, *nverr; 624204076Spjd const char *resname; 625204076Spjd const unsigned char *token; 626204076Spjd char laddr[256], raddr[256]; 627204076Spjd size_t size; 628204076Spjd pid_t pid; 629204076Spjd int status; 630204076Spjd 631204076Spjd proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr)); 632204076Spjd pjdlog_debug(1, "Accepting connection to %s.", laddr); 633204076Spjd 634204076Spjd if (proto_accept(cfg->hc_listenconn, &conn) < 0) { 635204076Spjd pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 636204076Spjd return; 637204076Spjd } 638204076Spjd 639204076Spjd proto_local_address(conn, laddr, sizeof(laddr)); 640204076Spjd proto_remote_address(conn, raddr, sizeof(raddr)); 641209185Spjd pjdlog_info("Connection from %s to %s.", raddr, laddr); 642204076Spjd 643207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 644207371Spjd if (proto_timeout(conn, HAST_TIMEOUT) < 0) 645207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 646207371Spjd 647204076Spjd nvin = nvout = nverr = NULL; 648204076Spjd 649204076Spjd /* 650204076Spjd * Before receiving any data see if remote host have access to any 651204076Spjd * resource. 652204076Spjd */ 653204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 654204076Spjd if (proto_address_match(conn, res->hr_remoteaddr)) 655204076Spjd break; 656204076Spjd } 657204076Spjd if (res == NULL) { 658204076Spjd pjdlog_error("Client %s isn't known.", raddr); 659204076Spjd goto close; 660204076Spjd } 661204076Spjd /* Ok, remote host can access at least one resource. */ 662204076Spjd 663204076Spjd if (hast_proto_recv_hdr(conn, &nvin) < 0) { 664204076Spjd pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 665204076Spjd raddr); 666204076Spjd goto close; 667204076Spjd } 668204076Spjd 669204076Spjd resname = nv_get_string(nvin, "resource"); 670204076Spjd if (resname == NULL) { 671204076Spjd pjdlog_error("No 'resource' field in the header received from %s.", 672204076Spjd raddr); 673204076Spjd goto close; 674204076Spjd } 675204076Spjd pjdlog_debug(2, "%s: resource=%s", raddr, resname); 676204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 677204076Spjd /* 678204076Spjd * NULL token means that this is first conection. 679204076Spjd */ 680204076Spjd if (token != NULL && size != sizeof(res->hr_token)) { 681204076Spjd pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 682204076Spjd raddr, sizeof(res->hr_token), size); 683204076Spjd goto close; 684204076Spjd } 685204076Spjd 686204076Spjd /* 687204076Spjd * From now on we want to send errors to the remote node. 688204076Spjd */ 689204076Spjd nverr = nv_alloc(); 690204076Spjd 691204076Spjd /* Find resource related to this connection. */ 692204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 693204076Spjd if (strcmp(resname, res->hr_name) == 0) 694204076Spjd break; 695204076Spjd } 696204076Spjd /* Have we found the resource? */ 697204076Spjd if (res == NULL) { 698204076Spjd pjdlog_error("No resource '%s' as requested by %s.", 699204076Spjd resname, raddr); 700204076Spjd nv_add_stringf(nverr, "errmsg", "Resource not configured."); 701204076Spjd goto fail; 702204076Spjd } 703204076Spjd 704204076Spjd /* Now that we know resource name setup log prefix. */ 705204076Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 706204076Spjd 707204076Spjd /* Does the remote host have access to this resource? */ 708204076Spjd if (!proto_address_match(conn, res->hr_remoteaddr)) { 709204076Spjd pjdlog_error("Client %s has no access to the resource.", raddr); 710204076Spjd nv_add_stringf(nverr, "errmsg", "No access to the resource."); 711204076Spjd goto fail; 712204076Spjd } 713204076Spjd /* Is the resource marked as secondary? */ 714204076Spjd if (res->hr_role != HAST_ROLE_SECONDARY) { 715204076Spjd pjdlog_error("We act as %s for the resource and not as %s as requested by %s.", 716204076Spjd role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 717204076Spjd raddr); 718204076Spjd nv_add_stringf(nverr, "errmsg", 719204076Spjd "Remote node acts as %s for the resource and not as %s.", 720204076Spjd role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 721204076Spjd goto fail; 722204076Spjd } 723204076Spjd /* Does token (if exists) match? */ 724204076Spjd if (token != NULL && memcmp(token, res->hr_token, 725204076Spjd sizeof(res->hr_token)) != 0) { 726204076Spjd pjdlog_error("Token received from %s doesn't match.", raddr); 727209185Spjd nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 728204076Spjd goto fail; 729204076Spjd } 730204076Spjd /* 731204076Spjd * If there is no token, but we have half-open connection 732204076Spjd * (only remotein) or full connection (worker process is running) 733204076Spjd * we have to cancel those and accept the new connection. 734204076Spjd */ 735204076Spjd if (token == NULL) { 736218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 737204076Spjd pjdlog_debug(1, "Initial connection from %s.", raddr); 738204076Spjd if (res->hr_workerpid != 0) { 739218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 740204076Spjd pjdlog_debug(1, 741204076Spjd "Worker process exists (pid=%u), stopping it.", 742204076Spjd (unsigned int)res->hr_workerpid); 743204076Spjd /* Stop child process. */ 744204076Spjd if (kill(res->hr_workerpid, SIGINT) < 0) { 745204076Spjd pjdlog_errno(LOG_ERR, 746204076Spjd "Unable to stop worker process (pid=%u)", 747204076Spjd (unsigned int)res->hr_workerpid); 748204076Spjd /* 749204076Spjd * Other than logging the problem we 750204076Spjd * ignore it - nothing smart to do. 751204076Spjd */ 752204076Spjd } 753204076Spjd /* Wait for it to exit. */ 754204076Spjd else if ((pid = waitpid(res->hr_workerpid, 755204076Spjd &status, 0)) != res->hr_workerpid) { 756207372Spjd /* We can only log the problem. */ 757204076Spjd pjdlog_errno(LOG_ERR, 758204076Spjd "Waiting for worker process (pid=%u) failed", 759204076Spjd (unsigned int)res->hr_workerpid); 760204076Spjd } else { 761207372Spjd child_exit_log(res->hr_workerpid, status); 762204076Spjd } 763213006Spjd child_cleanup(res); 764204076Spjd } else if (res->hr_remotein != NULL) { 765204076Spjd char oaddr[256]; 766204076Spjd 767213981Spjd proto_remote_address(res->hr_remotein, oaddr, 768213981Spjd sizeof(oaddr)); 769204076Spjd pjdlog_debug(1, 770204076Spjd "Canceling half-open connection from %s on connection from %s.", 771204076Spjd oaddr, raddr); 772204076Spjd proto_close(res->hr_remotein); 773204076Spjd res->hr_remotein = NULL; 774204076Spjd } 775204076Spjd } 776204076Spjd 777204076Spjd /* 778204076Spjd * Checks and cleanups are done. 779204076Spjd */ 780204076Spjd 781204076Spjd if (token == NULL) { 782204076Spjd arc4random_buf(res->hr_token, sizeof(res->hr_token)); 783204076Spjd nvout = nv_alloc(); 784204076Spjd nv_add_uint8_array(nvout, res->hr_token, 785204076Spjd sizeof(res->hr_token), "token"); 786204076Spjd if (nv_error(nvout) != 0) { 787204076Spjd pjdlog_common(LOG_ERR, 0, nv_error(nvout), 788204076Spjd "Unable to prepare return header for %s", raddr); 789204076Spjd nv_add_stringf(nverr, "errmsg", 790204076Spjd "Remote node was unable to prepare return header: %s.", 791204076Spjd strerror(nv_error(nvout))); 792204076Spjd goto fail; 793204076Spjd } 794204076Spjd if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) { 795204076Spjd int error = errno; 796204076Spjd 797204076Spjd pjdlog_errno(LOG_ERR, "Unable to send response to %s", 798204076Spjd raddr); 799204076Spjd nv_add_stringf(nverr, "errmsg", 800204076Spjd "Remote node was unable to send response: %s.", 801204076Spjd strerror(error)); 802204076Spjd goto fail; 803204076Spjd } 804204076Spjd res->hr_remotein = conn; 805204076Spjd pjdlog_debug(1, "Incoming connection from %s configured.", 806204076Spjd raddr); 807204076Spjd } else { 808204076Spjd res->hr_remoteout = conn; 809204076Spjd pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 810204076Spjd hastd_secondary(res, nvin); 811204076Spjd } 812204076Spjd nv_free(nvin); 813204076Spjd nv_free(nvout); 814204076Spjd nv_free(nverr); 815204076Spjd pjdlog_prefix_set("%s", ""); 816204076Spjd return; 817204076Spjdfail: 818204076Spjd if (nv_error(nverr) != 0) { 819204076Spjd pjdlog_common(LOG_ERR, 0, nv_error(nverr), 820204076Spjd "Unable to prepare error header for %s", raddr); 821204076Spjd goto close; 822204076Spjd } 823204076Spjd if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) { 824204076Spjd pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 825204076Spjd goto close; 826204076Spjd } 827204076Spjdclose: 828204076Spjd if (nvin != NULL) 829204076Spjd nv_free(nvin); 830204076Spjd if (nvout != NULL) 831204076Spjd nv_free(nvout); 832204076Spjd if (nverr != NULL) 833204076Spjd nv_free(nverr); 834204076Spjd proto_close(conn); 835204076Spjd pjdlog_prefix_set("%s", ""); 836204076Spjd} 837204076Spjd 838204076Spjdstatic void 839218218Spjdconnection_migrate(struct hast_resource *res) 840218218Spjd{ 841218218Spjd struct proto_conn *conn; 842218218Spjd int16_t val = 0; 843218218Spjd 844218218Spjd if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) { 845218218Spjd pjdlog_errno(LOG_WARNING, 846218218Spjd "Unable to receive connection command"); 847218218Spjd return; 848218218Spjd } 849218218Spjd if (proto_client(res->hr_remoteaddr, &conn) < 0) { 850218218Spjd val = errno; 851218218Spjd pjdlog_errno(LOG_WARNING, 852218218Spjd "Unable to create outgoing connection to %s", 853218218Spjd res->hr_remoteaddr); 854218218Spjd goto out; 855218218Spjd } 856218218Spjd if (proto_connect(conn, -1) < 0) { 857218218Spjd val = errno; 858218218Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 859218218Spjd res->hr_remoteaddr); 860218218Spjd proto_close(conn); 861218218Spjd goto out; 862218218Spjd } 863218218Spjd val = 0; 864218218Spjdout: 865218218Spjd if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) { 866218218Spjd pjdlog_errno(LOG_WARNING, 867218218Spjd "Unable to send reply to connection request"); 868218218Spjd } 869218218Spjd if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0) 870218218Spjd pjdlog_errno(LOG_WARNING, "Unable to send connection"); 871218218Spjd} 872218218Spjd 873218218Spjdstatic void 874204076Spjdmain_loop(void) 875204076Spjd{ 876212038Spjd struct hast_resource *res; 877213009Spjd struct timeval seltimeout; 878213009Spjd struct timespec sigtimeout; 879213009Spjd int fd, maxfd, ret, signo; 880213009Spjd sigset_t mask; 881212037Spjd fd_set rfds; 882204076Spjd 883213009Spjd seltimeout.tv_sec = REPORT_INTERVAL; 884213009Spjd seltimeout.tv_usec = 0; 885213009Spjd sigtimeout.tv_sec = 0; 886213009Spjd sigtimeout.tv_nsec = 0; 887211977Spjd 888213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 889213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 890213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 891213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 892213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 893213009Spjd 894216477Spjd pjdlog_info("Started successfully, running protocol version %d.", 895216477Spjd HAST_PROTO_VERSION); 896216477Spjd 897204076Spjd for (;;) { 898213009Spjd while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 899213009Spjd switch (signo) { 900213009Spjd case SIGINT: 901213009Spjd case SIGTERM: 902213009Spjd sigexit_received = true; 903213009Spjd terminate_workers(); 904217967Spjd proto_close(cfg->hc_controlconn); 905213009Spjd exit(EX_OK); 906213009Spjd break; 907213009Spjd case SIGCHLD: 908213009Spjd child_exit(); 909213009Spjd break; 910213009Spjd case SIGHUP: 911213009Spjd hastd_reload(); 912213009Spjd break; 913213009Spjd default: 914218138Spjd PJDLOG_ABORT("Unexpected signal (%d).", signo); 915213009Spjd } 916211899Spjd } 917204076Spjd 918209177Spjd /* Setup descriptors for select(2). */ 919204076Spjd FD_ZERO(&rfds); 920212038Spjd maxfd = fd = proto_descriptor(cfg->hc_controlconn); 921218138Spjd PJDLOG_ASSERT(fd >= 0); 922212038Spjd FD_SET(fd, &rfds); 923212038Spjd fd = proto_descriptor(cfg->hc_listenconn); 924218138Spjd PJDLOG_ASSERT(fd >= 0); 925212038Spjd FD_SET(fd, &rfds); 926212038Spjd maxfd = fd > maxfd ? fd : maxfd; 927212038Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 928212038Spjd if (res->hr_event == NULL) 929212038Spjd continue; 930218218Spjd PJDLOG_ASSERT(res->hr_conn != NULL); 931212038Spjd fd = proto_descriptor(res->hr_event); 932218138Spjd PJDLOG_ASSERT(fd >= 0); 933212038Spjd FD_SET(fd, &rfds); 934212038Spjd maxfd = fd > maxfd ? fd : maxfd; 935218218Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 936218218Spjd /* Only primary workers asks for connections. */ 937218218Spjd fd = proto_descriptor(res->hr_conn); 938218218Spjd PJDLOG_ASSERT(fd >= 0); 939218218Spjd FD_SET(fd, &rfds); 940218218Spjd maxfd = fd > maxfd ? fd : maxfd; 941218218Spjd } 942212038Spjd } 943204076Spjd 944218138Spjd PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE); 945213009Spjd ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 946211977Spjd if (ret == 0) 947213429Spjd hook_check(); 948211977Spjd else if (ret == -1) { 949204076Spjd if (errno == EINTR) 950204076Spjd continue; 951204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 952204076Spjd pjdlog_exit(EX_OSERR, "select() failed"); 953204076Spjd } 954204076Spjd 955212038Spjd if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 956204076Spjd control_handle(cfg); 957212038Spjd if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds)) 958204076Spjd listen_accept(); 959212038Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 960212038Spjd if (res->hr_event == NULL) 961212038Spjd continue; 962218218Spjd PJDLOG_ASSERT(res->hr_conn != NULL); 963212038Spjd if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 964212038Spjd if (event_recv(res) == 0) 965212038Spjd continue; 966212038Spjd /* The worker process exited? */ 967212038Spjd proto_close(res->hr_event); 968212038Spjd res->hr_event = NULL; 969218218Spjd proto_close(res->hr_conn); 970218218Spjd res->hr_conn = NULL; 971218218Spjd continue; 972212038Spjd } 973218218Spjd if (res->hr_role == HAST_ROLE_PRIMARY && 974218218Spjd FD_ISSET(proto_descriptor(res->hr_conn), &rfds)) { 975218218Spjd connection_migrate(res); 976218218Spjd } 977212038Spjd } 978204076Spjd } 979204076Spjd} 980204076Spjd 981213428Spjdstatic void 982213428Spjddummy_sighandler(int sig __unused) 983213428Spjd{ 984213428Spjd /* Nothing to do. */ 985213428Spjd} 986213428Spjd 987204076Spjdint 988204076Spjdmain(int argc, char *argv[]) 989204076Spjd{ 990204076Spjd const char *pidfile; 991204076Spjd pid_t otherpid; 992204076Spjd bool foreground; 993204076Spjd int debuglevel; 994213009Spjd sigset_t mask; 995204076Spjd 996204076Spjd foreground = false; 997204076Spjd debuglevel = 0; 998204076Spjd pidfile = HASTD_PIDFILE; 999204076Spjd 1000204076Spjd for (;;) { 1001204076Spjd int ch; 1002204076Spjd 1003204076Spjd ch = getopt(argc, argv, "c:dFhP:"); 1004204076Spjd if (ch == -1) 1005204076Spjd break; 1006204076Spjd switch (ch) { 1007204076Spjd case 'c': 1008204076Spjd cfgpath = optarg; 1009204076Spjd break; 1010204076Spjd case 'd': 1011204076Spjd debuglevel++; 1012204076Spjd break; 1013204076Spjd case 'F': 1014204076Spjd foreground = true; 1015204076Spjd break; 1016204076Spjd case 'P': 1017204076Spjd pidfile = optarg; 1018204076Spjd break; 1019204076Spjd case 'h': 1020204076Spjd default: 1021204076Spjd usage(); 1022204076Spjd } 1023204076Spjd } 1024204076Spjd argc -= optind; 1025204076Spjd argv += optind; 1026204076Spjd 1027217965Spjd pjdlog_init(PJDLOG_MODE_STD); 1028204076Spjd pjdlog_debug_set(debuglevel); 1029204076Spjd 1030214273Spjd g_gate_load(); 1031214273Spjd 1032204076Spjd pfh = pidfile_open(pidfile, 0600, &otherpid); 1033204076Spjd if (pfh == NULL) { 1034204076Spjd if (errno == EEXIST) { 1035204076Spjd pjdlog_exitx(EX_TEMPFAIL, 1036204076Spjd "Another hastd is already running, pid: %jd.", 1037204076Spjd (intmax_t)otherpid); 1038204076Spjd } 1039204076Spjd /* If we cannot create pidfile from other reasons, only warn. */ 1040210879Spjd pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile"); 1041204076Spjd } 1042204076Spjd 1043210883Spjd cfg = yy_config_parse(cfgpath, true); 1044218138Spjd PJDLOG_ASSERT(cfg != NULL); 1045204076Spjd 1046213428Spjd /* 1047217307Spjd * Restore default actions for interesting signals in case parent 1048217307Spjd * process (like init(8)) decided to ignore some of them (like SIGHUP). 1049217307Spjd */ 1050217307Spjd PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 1051217307Spjd PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 1052217307Spjd PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 1053217307Spjd /* 1054213428Spjd * Because SIGCHLD is ignored by default, setup dummy handler for it, 1055213428Spjd * so we can mask it. 1056213428Spjd */ 1057213428Spjd PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 1058217307Spjd 1059213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1060213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1061213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1062213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1063213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1064213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 1065204076Spjd 1066204076Spjd /* Listen on control address. */ 1067204076Spjd if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) { 1068204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1069204076Spjd pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 1070204076Spjd cfg->hc_controladdr); 1071204076Spjd } 1072204076Spjd /* Listen for remote connections. */ 1073204076Spjd if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) { 1074204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1075204076Spjd pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 1076204076Spjd cfg->hc_listenaddr); 1077204076Spjd } 1078204076Spjd 1079204076Spjd if (!foreground) { 1080204076Spjd if (daemon(0, 0) < 0) { 1081204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1082204076Spjd pjdlog_exit(EX_OSERR, "Unable to daemonize"); 1083204076Spjd } 1084204076Spjd 1085204076Spjd /* Start logging to syslog. */ 1086204076Spjd pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 1087204076Spjd 1088204076Spjd /* Write PID to a file. */ 1089204076Spjd if (pidfile_write(pfh) < 0) { 1090204076Spjd pjdlog_errno(LOG_WARNING, 1091204076Spjd "Unable to write PID to a file"); 1092204076Spjd } 1093204076Spjd } 1094204076Spjd 1095211977Spjd hook_init(); 1096211977Spjd 1097204076Spjd main_loop(); 1098204076Spjd 1099204076Spjd exit(0); 1100204076Spjd} 1101