hastd.c revision 219814
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 219814 2011-03-21 08:33:58Z 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> 49219813Spjd#include <time.h> 50204076Spjd#include <unistd.h> 51204076Spjd 52204076Spjd#include <activemap.h> 53204076Spjd#include <pjdlog.h> 54204076Spjd 55204076Spjd#include "control.h" 56212038Spjd#include "event.h" 57204076Spjd#include "hast.h" 58204076Spjd#include "hast_proto.h" 59204076Spjd#include "hastd.h" 60211977Spjd#include "hooks.h" 61204076Spjd#include "subr.h" 62204076Spjd 63204076Spjd/* Path to configuration file. */ 64210886Spjdconst char *cfgpath = HAST_CONFIG; 65204076Spjd/* Hastd configuration. */ 66204076Spjdstatic struct hastd_config *cfg; 67204076Spjd/* Was SIGINT or SIGTERM signal received? */ 68204076Spjdbool sigexit_received = false; 69204076Spjd/* PID file handle. */ 70204076Spjdstruct pidfh *pfh; 71204076Spjd 72211977Spjd/* How often check for hooks running for too long. */ 73213430Spjd#define REPORT_INTERVAL 5 74211977Spjd 75204076Spjdstatic void 76204076Spjdusage(void) 77204076Spjd{ 78204076Spjd 79204076Spjd errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]"); 80204076Spjd} 81204076Spjd 82204076Spjdstatic void 83204076Spjdg_gate_load(void) 84204076Spjd{ 85204076Spjd 86204076Spjd if (modfind("g_gate") == -1) { 87204076Spjd /* Not present in kernel, try loading it. */ 88204076Spjd if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) { 89204076Spjd if (errno != EEXIST) { 90204076Spjd pjdlog_exit(EX_OSERR, 91204076Spjd "Unable to load geom_gate module"); 92204076Spjd } 93204076Spjd } 94204076Spjd } 95204076Spjd} 96204076Spjd 97218041Spjdvoid 98218041Spjddescriptors_cleanup(struct hast_resource *res) 99218041Spjd{ 100218041Spjd struct hast_resource *tres; 101218041Spjd 102218041Spjd TAILQ_FOREACH(tres, &cfg->hc_resources, hr_next) { 103218041Spjd if (tres == res) { 104218041Spjd PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY || 105218041Spjd (res->hr_remotein == NULL && 106218041Spjd res->hr_remoteout == NULL)); 107218041Spjd continue; 108218041Spjd } 109218041Spjd if (tres->hr_remotein != NULL) 110218041Spjd proto_close(tres->hr_remotein); 111218041Spjd if (tres->hr_remoteout != NULL) 112218041Spjd proto_close(tres->hr_remoteout); 113218370Spjd if (tres->hr_ctrl != NULL) 114218370Spjd proto_close(tres->hr_ctrl); 115218370Spjd if (tres->hr_event != NULL) 116218370Spjd proto_close(tres->hr_event); 117218370Spjd if (tres->hr_conn != NULL) 118218370Spjd proto_close(tres->hr_conn); 119218041Spjd } 120218041Spjd if (cfg->hc_controlin != NULL) 121218041Spjd proto_close(cfg->hc_controlin); 122218041Spjd proto_close(cfg->hc_controlconn); 123218041Spjd proto_close(cfg->hc_listenconn); 124218041Spjd (void)pidfile_close(pfh); 125218041Spjd hook_fini(); 126218041Spjd pjdlog_fini(); 127218041Spjd} 128218041Spjd 129218044Spjdstatic const char * 130218044Spjddtype2str(mode_t mode) 131218044Spjd{ 132218044Spjd 133218044Spjd if (S_ISBLK(mode)) 134218044Spjd return ("block device"); 135218044Spjd else if (S_ISCHR(mode)) 136218044Spjd return ("character device"); 137218044Spjd else if (S_ISDIR(mode)) 138218044Spjd return ("directory"); 139218044Spjd else if (S_ISFIFO(mode)) 140218044Spjd return ("pipe or FIFO"); 141218044Spjd else if (S_ISLNK(mode)) 142218044Spjd return ("symbolic link"); 143218044Spjd else if (S_ISREG(mode)) 144218044Spjd return ("regular file"); 145218044Spjd else if (S_ISSOCK(mode)) 146218044Spjd return ("socket"); 147218044Spjd else if (S_ISWHT(mode)) 148218044Spjd return ("whiteout"); 149218044Spjd else 150218044Spjd return ("unknown"); 151218044Spjd} 152218044Spjd 153218044Spjdvoid 154218044Spjddescriptors_assert(const struct hast_resource *res, int pjdlogmode) 155218044Spjd{ 156218044Spjd char msg[256]; 157218044Spjd struct stat sb; 158218044Spjd long maxfd; 159218044Spjd bool isopen; 160218044Spjd mode_t mode; 161218044Spjd int fd; 162218044Spjd 163218044Spjd /* 164218044Spjd * At this point descriptor to syslog socket is closed, so if we want 165218044Spjd * to log assertion message, we have to first store it in 'msg' local 166218044Spjd * buffer and then open syslog socket and log it. 167218044Spjd */ 168218044Spjd msg[0] = '\0'; 169218044Spjd 170218044Spjd maxfd = sysconf(_SC_OPEN_MAX); 171218044Spjd if (maxfd < 0) { 172218373Spjd pjdlog_init(pjdlogmode); 173218373Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 174218373Spjd role2str(res->hr_role)); 175218044Spjd pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed"); 176218373Spjd pjdlog_fini(); 177218044Spjd maxfd = 16384; 178218044Spjd } 179218044Spjd for (fd = 0; fd <= maxfd; fd++) { 180218044Spjd if (fstat(fd, &sb) == 0) { 181218044Spjd isopen = true; 182218044Spjd mode = sb.st_mode; 183218044Spjd } else if (errno == EBADF) { 184218044Spjd isopen = false; 185218044Spjd mode = 0; 186218044Spjd } else { 187218375Spjd (void)snprintf(msg, sizeof(msg), 188218044Spjd "Unable to fstat descriptor %d: %s", fd, 189218044Spjd strerror(errno)); 190218374Spjd break; 191218044Spjd } 192218044Spjd if (fd == STDIN_FILENO || fd == STDOUT_FILENO || 193218044Spjd fd == STDERR_FILENO) { 194218044Spjd if (!isopen) { 195218375Spjd (void)snprintf(msg, sizeof(msg), 196218044Spjd "Descriptor %d (%s) is closed, but should be open.", 197218044Spjd fd, (fd == STDIN_FILENO ? "stdin" : 198218044Spjd (fd == STDOUT_FILENO ? "stdout" : "stderr"))); 199218044Spjd break; 200218044Spjd } 201218044Spjd } else if (fd == proto_descriptor(res->hr_event)) { 202218044Spjd if (!isopen) { 203218375Spjd (void)snprintf(msg, sizeof(msg), 204218044Spjd "Descriptor %d (event) is closed, but should be open.", 205218044Spjd fd); 206218044Spjd break; 207218044Spjd } 208218044Spjd if (!S_ISSOCK(mode)) { 209218375Spjd (void)snprintf(msg, sizeof(msg), 210218044Spjd "Descriptor %d (event) is %s, but should be %s.", 211218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 212218044Spjd break; 213218044Spjd } 214218044Spjd } else if (fd == proto_descriptor(res->hr_ctrl)) { 215218044Spjd if (!isopen) { 216218375Spjd (void)snprintf(msg, sizeof(msg), 217218044Spjd "Descriptor %d (ctrl) is closed, but should be open.", 218218044Spjd fd); 219218044Spjd break; 220218044Spjd } 221218044Spjd if (!S_ISSOCK(mode)) { 222218375Spjd (void)snprintf(msg, sizeof(msg), 223218044Spjd "Descriptor %d (ctrl) is %s, but should be %s.", 224218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 225218044Spjd break; 226218044Spjd } 227218218Spjd } else if (fd == proto_descriptor(res->hr_conn)) { 228218218Spjd if (!isopen) { 229218375Spjd (void)snprintf(msg, sizeof(msg), 230218218Spjd "Descriptor %d (conn) is closed, but should be open.", 231218218Spjd fd); 232218218Spjd break; 233218218Spjd } 234218218Spjd if (!S_ISSOCK(mode)) { 235218375Spjd (void)snprintf(msg, sizeof(msg), 236218218Spjd "Descriptor %d (conn) is %s, but should be %s.", 237218218Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 238218218Spjd break; 239218218Spjd } 240218044Spjd } else if (res->hr_role == HAST_ROLE_SECONDARY && 241218044Spjd fd == proto_descriptor(res->hr_remotein)) { 242218044Spjd if (!isopen) { 243218375Spjd (void)snprintf(msg, sizeof(msg), 244218044Spjd "Descriptor %d (remote in) is closed, but should be open.", 245218044Spjd fd); 246218044Spjd break; 247218044Spjd } 248218044Spjd if (!S_ISSOCK(mode)) { 249218375Spjd (void)snprintf(msg, sizeof(msg), 250218044Spjd "Descriptor %d (remote in) is %s, but should be %s.", 251218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 252218044Spjd break; 253218044Spjd } 254218044Spjd } else if (res->hr_role == HAST_ROLE_SECONDARY && 255218044Spjd fd == proto_descriptor(res->hr_remoteout)) { 256218044Spjd if (!isopen) { 257218375Spjd (void)snprintf(msg, sizeof(msg), 258218044Spjd "Descriptor %d (remote out) is closed, but should be open.", 259218044Spjd fd); 260218044Spjd break; 261218044Spjd } 262218044Spjd if (!S_ISSOCK(mode)) { 263218375Spjd (void)snprintf(msg, sizeof(msg), 264218044Spjd "Descriptor %d (remote out) is %s, but should be %s.", 265218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 266218044Spjd break; 267218044Spjd } 268218044Spjd } else { 269218044Spjd if (isopen) { 270218375Spjd (void)snprintf(msg, sizeof(msg), 271218044Spjd "Descriptor %d is open (%s), but should be closed.", 272218044Spjd fd, dtype2str(mode)); 273218044Spjd break; 274218044Spjd } 275218044Spjd } 276218044Spjd } 277218044Spjd if (msg[0] != '\0') { 278218044Spjd pjdlog_init(pjdlogmode); 279218044Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 280218044Spjd role2str(res->hr_role)); 281218044Spjd PJDLOG_ABORT("%s", msg); 282218044Spjd } 283218044Spjd} 284218044Spjd 285204076Spjdstatic void 286207372Spjdchild_exit_log(unsigned int pid, int status) 287207372Spjd{ 288207372Spjd 289207372Spjd if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 290207372Spjd pjdlog_debug(1, "Worker process exited gracefully (pid=%u).", 291207372Spjd pid); 292207372Spjd } else if (WIFSIGNALED(status)) { 293207372Spjd pjdlog_error("Worker process killed (pid=%u, signal=%d).", 294207372Spjd pid, WTERMSIG(status)); 295207372Spjd } else { 296207372Spjd pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).", 297207372Spjd pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1); 298207372Spjd } 299207372Spjd} 300207372Spjd 301207372Spjdstatic void 302204076Spjdchild_exit(void) 303204076Spjd{ 304204076Spjd struct hast_resource *res; 305204076Spjd int status; 306204076Spjd pid_t pid; 307204076Spjd 308204076Spjd while ((pid = wait3(&status, WNOHANG, NULL)) > 0) { 309204076Spjd /* Find resource related to the process that just exited. */ 310204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 311204076Spjd if (pid == res->hr_workerpid) 312204076Spjd break; 313204076Spjd } 314204076Spjd if (res == NULL) { 315204076Spjd /* 316204076Spjd * This can happen when new connection arrives and we 317211977Spjd * cancel child responsible for the old one or if this 318211977Spjd * was hook which we executed. 319204076Spjd */ 320211977Spjd hook_check_one(pid, status); 321204076Spjd continue; 322204076Spjd } 323204076Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 324204076Spjd role2str(res->hr_role)); 325207372Spjd child_exit_log(pid, status); 326213006Spjd child_cleanup(res); 327204076Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 328207372Spjd /* 329207372Spjd * Restart child process if it was killed by signal 330207372Spjd * or exited because of temporary problem. 331207372Spjd */ 332207372Spjd if (WIFSIGNALED(status) || 333207372Spjd (WIFEXITED(status) && 334207372Spjd WEXITSTATUS(status) == EX_TEMPFAIL)) { 335207348Spjd sleep(1); 336207348Spjd pjdlog_info("Restarting worker process."); 337207348Spjd hastd_primary(res); 338207348Spjd } else { 339207348Spjd res->hr_role = HAST_ROLE_INIT; 340207348Spjd pjdlog_info("Changing resource role back to %s.", 341207348Spjd role2str(res->hr_role)); 342207348Spjd } 343204076Spjd } 344204076Spjd pjdlog_prefix_set("%s", ""); 345204076Spjd } 346204076Spjd} 347204076Spjd 348210886Spjdstatic bool 349210886Spjdresource_needs_restart(const struct hast_resource *res0, 350210886Spjd const struct hast_resource *res1) 351210886Spjd{ 352210886Spjd 353218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 354210886Spjd 355210886Spjd if (strcmp(res0->hr_provname, res1->hr_provname) != 0) 356210886Spjd return (true); 357210886Spjd if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0) 358210886Spjd return (true); 359210886Spjd if (res0->hr_role == HAST_ROLE_INIT || 360210886Spjd res0->hr_role == HAST_ROLE_SECONDARY) { 361210886Spjd if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 362210886Spjd return (true); 363210886Spjd if (res0->hr_replication != res1->hr_replication) 364210886Spjd return (true); 365219351Spjd if (res0->hr_checksum != res1->hr_checksum) 366219351Spjd return (true); 367219354Spjd if (res0->hr_compression != res1->hr_compression) 368219354Spjd return (true); 369210886Spjd if (res0->hr_timeout != res1->hr_timeout) 370210886Spjd return (true); 371211886Spjd if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 372211886Spjd return (true); 373210886Spjd } 374210886Spjd return (false); 375210886Spjd} 376210886Spjd 377210886Spjdstatic bool 378210886Spjdresource_needs_reload(const struct hast_resource *res0, 379210886Spjd const struct hast_resource *res1) 380210886Spjd{ 381210886Spjd 382218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 383218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0); 384218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0); 385210886Spjd 386210886Spjd if (res0->hr_role != HAST_ROLE_PRIMARY) 387210886Spjd return (false); 388210886Spjd 389210886Spjd if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 390210886Spjd return (true); 391210886Spjd if (res0->hr_replication != res1->hr_replication) 392210886Spjd return (true); 393219351Spjd if (res0->hr_checksum != res1->hr_checksum) 394219351Spjd return (true); 395219354Spjd if (res0->hr_compression != res1->hr_compression) 396219354Spjd return (true); 397210886Spjd if (res0->hr_timeout != res1->hr_timeout) 398210886Spjd return (true); 399211886Spjd if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 400211886Spjd return (true); 401210886Spjd return (false); 402210886Spjd} 403210886Spjd 404204076Spjdstatic void 405217784Spjdresource_reload(const struct hast_resource *res) 406217784Spjd{ 407217784Spjd struct nv *nvin, *nvout; 408217784Spjd int error; 409217784Spjd 410218138Spjd PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 411217784Spjd 412217784Spjd nvout = nv_alloc(); 413217784Spjd nv_add_uint8(nvout, HASTCTL_RELOAD, "cmd"); 414217784Spjd nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr"); 415217784Spjd nv_add_int32(nvout, (int32_t)res->hr_replication, "replication"); 416219351Spjd nv_add_int32(nvout, (int32_t)res->hr_checksum, "checksum"); 417219354Spjd nv_add_int32(nvout, (int32_t)res->hr_compression, "compression"); 418217784Spjd nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout"); 419217784Spjd nv_add_string(nvout, res->hr_exec, "exec"); 420217784Spjd if (nv_error(nvout) != 0) { 421217784Spjd nv_free(nvout); 422217784Spjd pjdlog_error("Unable to allocate header for reload message."); 423217784Spjd return; 424217784Spjd } 425217784Spjd if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) { 426217784Spjd pjdlog_errno(LOG_ERR, "Unable to send reload message"); 427217784Spjd nv_free(nvout); 428217784Spjd return; 429217784Spjd } 430217784Spjd nv_free(nvout); 431217784Spjd 432217784Spjd /* Receive response. */ 433217784Spjd if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) { 434217784Spjd pjdlog_errno(LOG_ERR, "Unable to receive reload reply"); 435217784Spjd return; 436217784Spjd } 437217784Spjd error = nv_get_int16(nvin, "error"); 438217784Spjd nv_free(nvin); 439217784Spjd if (error != 0) { 440217784Spjd pjdlog_common(LOG_ERR, 0, error, "Reload failed"); 441217784Spjd return; 442217784Spjd } 443217784Spjd} 444217784Spjd 445217784Spjdstatic void 446204076Spjdhastd_reload(void) 447204076Spjd{ 448210886Spjd struct hastd_config *newcfg; 449210886Spjd struct hast_resource *nres, *cres, *tres; 450210886Spjd uint8_t role; 451204076Spjd 452210886Spjd pjdlog_info("Reloading configuration..."); 453210886Spjd 454210886Spjd newcfg = yy_config_parse(cfgpath, false); 455210886Spjd if (newcfg == NULL) 456210886Spjd goto failed; 457210886Spjd 458210886Spjd /* 459210886Spjd * Check if control address has changed. 460210886Spjd */ 461210886Spjd if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 462210886Spjd if (proto_server(newcfg->hc_controladdr, 463210886Spjd &newcfg->hc_controlconn) < 0) { 464210886Spjd pjdlog_errno(LOG_ERR, 465210886Spjd "Unable to listen on control address %s", 466210886Spjd newcfg->hc_controladdr); 467210886Spjd goto failed; 468210886Spjd } 469210886Spjd } 470210886Spjd /* 471210886Spjd * Check if listen address has changed. 472210886Spjd */ 473210886Spjd if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) { 474210886Spjd if (proto_server(newcfg->hc_listenaddr, 475210886Spjd &newcfg->hc_listenconn) < 0) { 476210886Spjd pjdlog_errno(LOG_ERR, "Unable to listen on address %s", 477210886Spjd newcfg->hc_listenaddr); 478210886Spjd goto failed; 479210886Spjd } 480210886Spjd } 481210886Spjd /* 482210886Spjd * Only when both control and listen sockets are successfully 483210886Spjd * initialized switch them to new configuration. 484210886Spjd */ 485210886Spjd if (newcfg->hc_controlconn != NULL) { 486210886Spjd pjdlog_info("Control socket changed from %s to %s.", 487210886Spjd cfg->hc_controladdr, newcfg->hc_controladdr); 488210886Spjd proto_close(cfg->hc_controlconn); 489210886Spjd cfg->hc_controlconn = newcfg->hc_controlconn; 490210886Spjd newcfg->hc_controlconn = NULL; 491210886Spjd strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 492210886Spjd sizeof(cfg->hc_controladdr)); 493210886Spjd } 494210886Spjd if (newcfg->hc_listenconn != NULL) { 495210886Spjd pjdlog_info("Listen socket changed from %s to %s.", 496210886Spjd cfg->hc_listenaddr, newcfg->hc_listenaddr); 497210886Spjd proto_close(cfg->hc_listenconn); 498210886Spjd cfg->hc_listenconn = newcfg->hc_listenconn; 499210886Spjd newcfg->hc_listenconn = NULL; 500210886Spjd strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr, 501210886Spjd sizeof(cfg->hc_listenaddr)); 502210886Spjd } 503210886Spjd 504210886Spjd /* 505210886Spjd * Stop and remove resources that were removed from the configuration. 506210886Spjd */ 507210886Spjd TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 508210886Spjd TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 509210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 510210886Spjd break; 511210886Spjd } 512210886Spjd if (nres == NULL) { 513210886Spjd control_set_role(cres, HAST_ROLE_INIT); 514210886Spjd TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 515210886Spjd pjdlog_info("Resource %s removed.", cres->hr_name); 516210886Spjd free(cres); 517210886Spjd } 518210886Spjd } 519210886Spjd /* 520210886Spjd * Move new resources to the current configuration. 521210886Spjd */ 522210886Spjd TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 523210886Spjd TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 524210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 525210886Spjd break; 526210886Spjd } 527210886Spjd if (cres == NULL) { 528210886Spjd TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 529210886Spjd TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 530210886Spjd pjdlog_info("Resource %s added.", nres->hr_name); 531210886Spjd } 532210886Spjd } 533210886Spjd /* 534210886Spjd * Deal with modified resources. 535210886Spjd * Depending on what has changed exactly we might want to perform 536210886Spjd * different actions. 537210886Spjd * 538210886Spjd * We do full resource restart in the following situations: 539210886Spjd * Resource role is INIT or SECONDARY. 540210886Spjd * Resource role is PRIMARY and path to local component or provider 541210886Spjd * name has changed. 542210886Spjd * In case of PRIMARY, the worker process will be killed and restarted, 543210886Spjd * which also means removing /dev/hast/<name> provider and 544210886Spjd * recreating it. 545210886Spjd * 546210886Spjd * We do just reload (send SIGHUP to worker process) if we act as 547217729Spjd * PRIMARY, but only if remote address, replication mode, timeout or 548217729Spjd * execution path has changed. For those, there is no need to restart 549217729Spjd * worker process. 550210886Spjd * If PRIMARY receives SIGHUP, it will reconnect if remote address or 551210886Spjd * replication mode has changed or simply set new timeout if only 552210886Spjd * timeout has changed. 553210886Spjd */ 554210886Spjd TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 555210886Spjd TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 556210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 557210886Spjd break; 558210886Spjd } 559218138Spjd PJDLOG_ASSERT(cres != NULL); 560210886Spjd if (resource_needs_restart(cres, nres)) { 561210886Spjd pjdlog_info("Resource %s configuration was modified, restarting it.", 562210886Spjd cres->hr_name); 563210886Spjd role = cres->hr_role; 564210886Spjd control_set_role(cres, HAST_ROLE_INIT); 565210886Spjd TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 566210886Spjd free(cres); 567210886Spjd TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 568210886Spjd TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 569210886Spjd control_set_role(nres, role); 570210886Spjd } else if (resource_needs_reload(cres, nres)) { 571210886Spjd pjdlog_info("Resource %s configuration was modified, reloading it.", 572210886Spjd cres->hr_name); 573210886Spjd strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 574210886Spjd sizeof(cres->hr_remoteaddr)); 575210886Spjd cres->hr_replication = nres->hr_replication; 576219351Spjd cres->hr_checksum = nres->hr_checksum; 577219354Spjd cres->hr_compression = nres->hr_compression; 578210886Spjd cres->hr_timeout = nres->hr_timeout; 579217729Spjd strlcpy(cres->hr_exec, nres->hr_exec, 580217729Spjd sizeof(cres->hr_exec)); 581217784Spjd if (cres->hr_workerpid != 0) 582217784Spjd resource_reload(cres); 583210886Spjd } 584210886Spjd } 585210886Spjd 586210886Spjd yy_config_free(newcfg); 587210886Spjd pjdlog_info("Configuration reloaded successfully."); 588210886Spjd return; 589210886Spjdfailed: 590210886Spjd if (newcfg != NULL) { 591210886Spjd if (newcfg->hc_controlconn != NULL) 592210886Spjd proto_close(newcfg->hc_controlconn); 593210886Spjd if (newcfg->hc_listenconn != NULL) 594210886Spjd proto_close(newcfg->hc_listenconn); 595210886Spjd yy_config_free(newcfg); 596210886Spjd } 597210886Spjd pjdlog_warning("Configuration not reloaded."); 598204076Spjd} 599204076Spjd 600204076Spjdstatic void 601211899Spjdterminate_workers(void) 602211899Spjd{ 603211899Spjd struct hast_resource *res; 604211899Spjd 605211899Spjd pjdlog_info("Termination signal received, exiting."); 606211899Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 607211899Spjd if (res->hr_workerpid == 0) 608211899Spjd continue; 609211899Spjd pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 610211899Spjd res->hr_name, role2str(res->hr_role), res->hr_workerpid); 611211899Spjd if (kill(res->hr_workerpid, SIGTERM) == 0) 612211899Spjd continue; 613211899Spjd pjdlog_errno(LOG_WARNING, 614211899Spjd "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 615211899Spjd res->hr_name, role2str(res->hr_role), res->hr_workerpid); 616211899Spjd } 617211899Spjd} 618211899Spjd 619211899Spjdstatic void 620204076Spjdlisten_accept(void) 621204076Spjd{ 622204076Spjd struct hast_resource *res; 623204076Spjd struct proto_conn *conn; 624204076Spjd struct nv *nvin, *nvout, *nverr; 625204076Spjd const char *resname; 626204076Spjd const unsigned char *token; 627204076Spjd char laddr[256], raddr[256]; 628204076Spjd size_t size; 629204076Spjd pid_t pid; 630204076Spjd int status; 631204076Spjd 632204076Spjd proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr)); 633204076Spjd pjdlog_debug(1, "Accepting connection to %s.", laddr); 634204076Spjd 635204076Spjd if (proto_accept(cfg->hc_listenconn, &conn) < 0) { 636204076Spjd pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 637204076Spjd return; 638204076Spjd } 639204076Spjd 640204076Spjd proto_local_address(conn, laddr, sizeof(laddr)); 641204076Spjd proto_remote_address(conn, raddr, sizeof(raddr)); 642209185Spjd pjdlog_info("Connection from %s to %s.", raddr, laddr); 643204076Spjd 644207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 645207371Spjd if (proto_timeout(conn, HAST_TIMEOUT) < 0) 646207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 647207371Spjd 648204076Spjd nvin = nvout = nverr = NULL; 649204076Spjd 650204076Spjd /* 651204076Spjd * Before receiving any data see if remote host have access to any 652204076Spjd * resource. 653204076Spjd */ 654204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 655204076Spjd if (proto_address_match(conn, res->hr_remoteaddr)) 656204076Spjd break; 657204076Spjd } 658204076Spjd if (res == NULL) { 659204076Spjd pjdlog_error("Client %s isn't known.", raddr); 660204076Spjd goto close; 661204076Spjd } 662204076Spjd /* Ok, remote host can access at least one resource. */ 663204076Spjd 664204076Spjd if (hast_proto_recv_hdr(conn, &nvin) < 0) { 665204076Spjd pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 666204076Spjd raddr); 667204076Spjd goto close; 668204076Spjd } 669204076Spjd 670204076Spjd resname = nv_get_string(nvin, "resource"); 671204076Spjd if (resname == NULL) { 672204076Spjd pjdlog_error("No 'resource' field in the header received from %s.", 673204076Spjd raddr); 674204076Spjd goto close; 675204076Spjd } 676204076Spjd pjdlog_debug(2, "%s: resource=%s", raddr, resname); 677204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 678204076Spjd /* 679204076Spjd * NULL token means that this is first conection. 680204076Spjd */ 681204076Spjd if (token != NULL && size != sizeof(res->hr_token)) { 682204076Spjd pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 683204076Spjd raddr, sizeof(res->hr_token), size); 684204076Spjd goto close; 685204076Spjd } 686204076Spjd 687204076Spjd /* 688204076Spjd * From now on we want to send errors to the remote node. 689204076Spjd */ 690204076Spjd nverr = nv_alloc(); 691204076Spjd 692204076Spjd /* Find resource related to this connection. */ 693204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 694204076Spjd if (strcmp(resname, res->hr_name) == 0) 695204076Spjd break; 696204076Spjd } 697204076Spjd /* Have we found the resource? */ 698204076Spjd if (res == NULL) { 699204076Spjd pjdlog_error("No resource '%s' as requested by %s.", 700204076Spjd resname, raddr); 701204076Spjd nv_add_stringf(nverr, "errmsg", "Resource not configured."); 702204076Spjd goto fail; 703204076Spjd } 704204076Spjd 705204076Spjd /* Now that we know resource name setup log prefix. */ 706204076Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 707204076Spjd 708204076Spjd /* Does the remote host have access to this resource? */ 709204076Spjd if (!proto_address_match(conn, res->hr_remoteaddr)) { 710204076Spjd pjdlog_error("Client %s has no access to the resource.", raddr); 711204076Spjd nv_add_stringf(nverr, "errmsg", "No access to the resource."); 712204076Spjd goto fail; 713204076Spjd } 714204076Spjd /* Is the resource marked as secondary? */ 715204076Spjd if (res->hr_role != HAST_ROLE_SECONDARY) { 716204076Spjd pjdlog_error("We act as %s for the resource and not as %s as requested by %s.", 717204076Spjd role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 718204076Spjd raddr); 719204076Spjd nv_add_stringf(nverr, "errmsg", 720204076Spjd "Remote node acts as %s for the resource and not as %s.", 721204076Spjd role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 722204076Spjd goto fail; 723204076Spjd } 724204076Spjd /* Does token (if exists) match? */ 725204076Spjd if (token != NULL && memcmp(token, res->hr_token, 726204076Spjd sizeof(res->hr_token)) != 0) { 727204076Spjd pjdlog_error("Token received from %s doesn't match.", raddr); 728209185Spjd nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 729204076Spjd goto fail; 730204076Spjd } 731204076Spjd /* 732204076Spjd * If there is no token, but we have half-open connection 733204076Spjd * (only remotein) or full connection (worker process is running) 734204076Spjd * we have to cancel those and accept the new connection. 735204076Spjd */ 736204076Spjd if (token == NULL) { 737218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 738204076Spjd pjdlog_debug(1, "Initial connection from %s.", raddr); 739204076Spjd if (res->hr_workerpid != 0) { 740218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 741204076Spjd pjdlog_debug(1, 742204076Spjd "Worker process exists (pid=%u), stopping it.", 743204076Spjd (unsigned int)res->hr_workerpid); 744204076Spjd /* Stop child process. */ 745204076Spjd if (kill(res->hr_workerpid, SIGINT) < 0) { 746204076Spjd pjdlog_errno(LOG_ERR, 747204076Spjd "Unable to stop worker process (pid=%u)", 748204076Spjd (unsigned int)res->hr_workerpid); 749204076Spjd /* 750204076Spjd * Other than logging the problem we 751204076Spjd * ignore it - nothing smart to do. 752204076Spjd */ 753204076Spjd } 754204076Spjd /* Wait for it to exit. */ 755204076Spjd else if ((pid = waitpid(res->hr_workerpid, 756204076Spjd &status, 0)) != res->hr_workerpid) { 757207372Spjd /* We can only log the problem. */ 758204076Spjd pjdlog_errno(LOG_ERR, 759204076Spjd "Waiting for worker process (pid=%u) failed", 760204076Spjd (unsigned int)res->hr_workerpid); 761204076Spjd } else { 762207372Spjd child_exit_log(res->hr_workerpid, status); 763204076Spjd } 764213006Spjd child_cleanup(res); 765204076Spjd } else if (res->hr_remotein != NULL) { 766204076Spjd char oaddr[256]; 767204076Spjd 768213981Spjd proto_remote_address(res->hr_remotein, oaddr, 769213981Spjd sizeof(oaddr)); 770204076Spjd pjdlog_debug(1, 771204076Spjd "Canceling half-open connection from %s on connection from %s.", 772204076Spjd oaddr, raddr); 773204076Spjd proto_close(res->hr_remotein); 774204076Spjd res->hr_remotein = NULL; 775204076Spjd } 776204076Spjd } 777204076Spjd 778204076Spjd /* 779204076Spjd * Checks and cleanups are done. 780204076Spjd */ 781204076Spjd 782204076Spjd if (token == NULL) { 783204076Spjd arc4random_buf(res->hr_token, sizeof(res->hr_token)); 784204076Spjd nvout = nv_alloc(); 785204076Spjd nv_add_uint8_array(nvout, res->hr_token, 786204076Spjd sizeof(res->hr_token), "token"); 787204076Spjd if (nv_error(nvout) != 0) { 788204076Spjd pjdlog_common(LOG_ERR, 0, nv_error(nvout), 789204076Spjd "Unable to prepare return header for %s", raddr); 790204076Spjd nv_add_stringf(nverr, "errmsg", 791204076Spjd "Remote node was unable to prepare return header: %s.", 792204076Spjd strerror(nv_error(nvout))); 793204076Spjd goto fail; 794204076Spjd } 795204076Spjd if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) { 796204076Spjd int error = errno; 797204076Spjd 798204076Spjd pjdlog_errno(LOG_ERR, "Unable to send response to %s", 799204076Spjd raddr); 800204076Spjd nv_add_stringf(nverr, "errmsg", 801204076Spjd "Remote node was unable to send response: %s.", 802204076Spjd strerror(error)); 803204076Spjd goto fail; 804204076Spjd } 805204076Spjd res->hr_remotein = conn; 806204076Spjd pjdlog_debug(1, "Incoming connection from %s configured.", 807204076Spjd raddr); 808204076Spjd } else { 809204076Spjd res->hr_remoteout = conn; 810204076Spjd pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 811204076Spjd hastd_secondary(res, nvin); 812204076Spjd } 813204076Spjd nv_free(nvin); 814204076Spjd nv_free(nvout); 815204076Spjd nv_free(nverr); 816204076Spjd pjdlog_prefix_set("%s", ""); 817204076Spjd return; 818204076Spjdfail: 819204076Spjd if (nv_error(nverr) != 0) { 820204076Spjd pjdlog_common(LOG_ERR, 0, nv_error(nverr), 821204076Spjd "Unable to prepare error header for %s", raddr); 822204076Spjd goto close; 823204076Spjd } 824204076Spjd if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) { 825204076Spjd pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 826204076Spjd goto close; 827204076Spjd } 828204076Spjdclose: 829204076Spjd if (nvin != NULL) 830204076Spjd nv_free(nvin); 831204076Spjd if (nvout != NULL) 832204076Spjd nv_free(nvout); 833204076Spjd if (nverr != NULL) 834204076Spjd nv_free(nverr); 835204076Spjd proto_close(conn); 836204076Spjd pjdlog_prefix_set("%s", ""); 837204076Spjd} 838204076Spjd 839204076Spjdstatic void 840218218Spjdconnection_migrate(struct hast_resource *res) 841218218Spjd{ 842218218Spjd struct proto_conn *conn; 843218218Spjd int16_t val = 0; 844218218Spjd 845219814Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 846219814Spjd 847218218Spjd if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) { 848218218Spjd pjdlog_errno(LOG_WARNING, 849218218Spjd "Unable to receive connection command"); 850218218Spjd return; 851218218Spjd } 852218218Spjd if (proto_client(res->hr_remoteaddr, &conn) < 0) { 853218218Spjd val = errno; 854218218Spjd pjdlog_errno(LOG_WARNING, 855218218Spjd "Unable to create outgoing connection to %s", 856218218Spjd res->hr_remoteaddr); 857218218Spjd goto out; 858218218Spjd } 859218218Spjd if (proto_connect(conn, -1) < 0) { 860218218Spjd val = errno; 861218218Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 862218218Spjd res->hr_remoteaddr); 863218218Spjd proto_close(conn); 864218218Spjd goto out; 865218218Spjd } 866218218Spjd val = 0; 867218218Spjdout: 868218218Spjd if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) { 869218218Spjd pjdlog_errno(LOG_WARNING, 870218218Spjd "Unable to send reply to connection request"); 871218218Spjd } 872218218Spjd if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0) 873218218Spjd pjdlog_errno(LOG_WARNING, "Unable to send connection"); 874219814Spjd 875219814Spjd pjdlog_prefix_set("%s", ""); 876218218Spjd} 877218218Spjd 878218218Spjdstatic void 879204076Spjdmain_loop(void) 880204076Spjd{ 881212038Spjd struct hast_resource *res; 882213009Spjd struct timeval seltimeout; 883213009Spjd struct timespec sigtimeout; 884213009Spjd int fd, maxfd, ret, signo; 885219813Spjd time_t lastcheck, now; 886213009Spjd sigset_t mask; 887212037Spjd fd_set rfds; 888204076Spjd 889219813Spjd lastcheck = time(NULL); 890213009Spjd seltimeout.tv_sec = REPORT_INTERVAL; 891213009Spjd seltimeout.tv_usec = 0; 892213009Spjd sigtimeout.tv_sec = 0; 893213009Spjd sigtimeout.tv_nsec = 0; 894211977Spjd 895213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 896213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 897213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 898213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 899213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 900213009Spjd 901216477Spjd pjdlog_info("Started successfully, running protocol version %d.", 902216477Spjd HAST_PROTO_VERSION); 903216477Spjd 904204076Spjd for (;;) { 905213009Spjd while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 906213009Spjd switch (signo) { 907213009Spjd case SIGINT: 908213009Spjd case SIGTERM: 909213009Spjd sigexit_received = true; 910213009Spjd terminate_workers(); 911217967Spjd proto_close(cfg->hc_controlconn); 912213009Spjd exit(EX_OK); 913213009Spjd break; 914213009Spjd case SIGCHLD: 915213009Spjd child_exit(); 916213009Spjd break; 917213009Spjd case SIGHUP: 918213009Spjd hastd_reload(); 919213009Spjd break; 920213009Spjd default: 921218138Spjd PJDLOG_ABORT("Unexpected signal (%d).", signo); 922213009Spjd } 923211899Spjd } 924204076Spjd 925209177Spjd /* Setup descriptors for select(2). */ 926204076Spjd FD_ZERO(&rfds); 927212038Spjd maxfd = fd = proto_descriptor(cfg->hc_controlconn); 928218138Spjd PJDLOG_ASSERT(fd >= 0); 929212038Spjd FD_SET(fd, &rfds); 930212038Spjd fd = proto_descriptor(cfg->hc_listenconn); 931218138Spjd PJDLOG_ASSERT(fd >= 0); 932212038Spjd FD_SET(fd, &rfds); 933212038Spjd maxfd = fd > maxfd ? fd : maxfd; 934212038Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 935212038Spjd if (res->hr_event == NULL) 936212038Spjd continue; 937218218Spjd PJDLOG_ASSERT(res->hr_conn != NULL); 938212038Spjd fd = proto_descriptor(res->hr_event); 939218138Spjd PJDLOG_ASSERT(fd >= 0); 940212038Spjd FD_SET(fd, &rfds); 941212038Spjd maxfd = fd > maxfd ? fd : maxfd; 942218218Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 943218218Spjd /* Only primary workers asks for connections. */ 944218218Spjd fd = proto_descriptor(res->hr_conn); 945218218Spjd PJDLOG_ASSERT(fd >= 0); 946218218Spjd FD_SET(fd, &rfds); 947218218Spjd maxfd = fd > maxfd ? fd : maxfd; 948218218Spjd } 949212038Spjd } 950204076Spjd 951218138Spjd PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE); 952213009Spjd ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 953219813Spjd now = time(NULL); 954219813Spjd if (lastcheck + REPORT_INTERVAL <= now) { 955213429Spjd hook_check(); 956219813Spjd lastcheck = now; 957219813Spjd } 958219813Spjd if (ret == 0) { 959219813Spjd /* 960219813Spjd * select(2) timed out, so there should be no 961219813Spjd * descriptors to check. 962219813Spjd */ 963219813Spjd continue; 964219813Spjd } else if (ret == -1) { 965204076Spjd if (errno == EINTR) 966204076Spjd continue; 967204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 968204076Spjd pjdlog_exit(EX_OSERR, "select() failed"); 969204076Spjd } 970204076Spjd 971212038Spjd if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 972204076Spjd control_handle(cfg); 973212038Spjd if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds)) 974204076Spjd listen_accept(); 975212038Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 976212038Spjd if (res->hr_event == NULL) 977212038Spjd continue; 978218218Spjd PJDLOG_ASSERT(res->hr_conn != NULL); 979212038Spjd if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 980212038Spjd if (event_recv(res) == 0) 981212038Spjd continue; 982212038Spjd /* The worker process exited? */ 983212038Spjd proto_close(res->hr_event); 984212038Spjd res->hr_event = NULL; 985218218Spjd proto_close(res->hr_conn); 986218218Spjd res->hr_conn = NULL; 987218218Spjd continue; 988212038Spjd } 989218218Spjd if (res->hr_role == HAST_ROLE_PRIMARY && 990218218Spjd FD_ISSET(proto_descriptor(res->hr_conn), &rfds)) { 991218218Spjd connection_migrate(res); 992218218Spjd } 993212038Spjd } 994204076Spjd } 995204076Spjd} 996204076Spjd 997213428Spjdstatic void 998213428Spjddummy_sighandler(int sig __unused) 999213428Spjd{ 1000213428Spjd /* Nothing to do. */ 1001213428Spjd} 1002213428Spjd 1003204076Spjdint 1004204076Spjdmain(int argc, char *argv[]) 1005204076Spjd{ 1006204076Spjd const char *pidfile; 1007204076Spjd pid_t otherpid; 1008204076Spjd bool foreground; 1009204076Spjd int debuglevel; 1010213009Spjd sigset_t mask; 1011204076Spjd 1012204076Spjd foreground = false; 1013204076Spjd debuglevel = 0; 1014204076Spjd pidfile = HASTD_PIDFILE; 1015204076Spjd 1016204076Spjd for (;;) { 1017204076Spjd int ch; 1018204076Spjd 1019204076Spjd ch = getopt(argc, argv, "c:dFhP:"); 1020204076Spjd if (ch == -1) 1021204076Spjd break; 1022204076Spjd switch (ch) { 1023204076Spjd case 'c': 1024204076Spjd cfgpath = optarg; 1025204076Spjd break; 1026204076Spjd case 'd': 1027204076Spjd debuglevel++; 1028204076Spjd break; 1029204076Spjd case 'F': 1030204076Spjd foreground = true; 1031204076Spjd break; 1032204076Spjd case 'P': 1033204076Spjd pidfile = optarg; 1034204076Spjd break; 1035204076Spjd case 'h': 1036204076Spjd default: 1037204076Spjd usage(); 1038204076Spjd } 1039204076Spjd } 1040204076Spjd argc -= optind; 1041204076Spjd argv += optind; 1042204076Spjd 1043217965Spjd pjdlog_init(PJDLOG_MODE_STD); 1044204076Spjd pjdlog_debug_set(debuglevel); 1045204076Spjd 1046214273Spjd g_gate_load(); 1047214273Spjd 1048204076Spjd pfh = pidfile_open(pidfile, 0600, &otherpid); 1049204076Spjd if (pfh == NULL) { 1050204076Spjd if (errno == EEXIST) { 1051204076Spjd pjdlog_exitx(EX_TEMPFAIL, 1052204076Spjd "Another hastd is already running, pid: %jd.", 1053204076Spjd (intmax_t)otherpid); 1054204076Spjd } 1055204076Spjd /* If we cannot create pidfile from other reasons, only warn. */ 1056210879Spjd pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile"); 1057204076Spjd } 1058204076Spjd 1059210883Spjd cfg = yy_config_parse(cfgpath, true); 1060218138Spjd PJDLOG_ASSERT(cfg != NULL); 1061204076Spjd 1062213428Spjd /* 1063217307Spjd * Restore default actions for interesting signals in case parent 1064217307Spjd * process (like init(8)) decided to ignore some of them (like SIGHUP). 1065217307Spjd */ 1066217307Spjd PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 1067217307Spjd PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 1068217307Spjd PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 1069217307Spjd /* 1070213428Spjd * Because SIGCHLD is ignored by default, setup dummy handler for it, 1071213428Spjd * so we can mask it. 1072213428Spjd */ 1073213428Spjd PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 1074217307Spjd 1075213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1076213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1077213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1078213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1079213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1080213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 1081204076Spjd 1082204076Spjd /* Listen on control address. */ 1083204076Spjd if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) { 1084204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1085204076Spjd pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 1086204076Spjd cfg->hc_controladdr); 1087204076Spjd } 1088204076Spjd /* Listen for remote connections. */ 1089204076Spjd if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) { 1090204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1091204076Spjd pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 1092204076Spjd cfg->hc_listenaddr); 1093204076Spjd } 1094204076Spjd 1095204076Spjd if (!foreground) { 1096204076Spjd if (daemon(0, 0) < 0) { 1097204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1098204076Spjd pjdlog_exit(EX_OSERR, "Unable to daemonize"); 1099204076Spjd } 1100204076Spjd 1101204076Spjd /* Start logging to syslog. */ 1102204076Spjd pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 1103204076Spjd 1104204076Spjd /* Write PID to a file. */ 1105204076Spjd if (pidfile_write(pfh) < 0) { 1106204076Spjd pjdlog_errno(LOG_WARNING, 1107204076Spjd "Unable to write PID to a file"); 1108204076Spjd } 1109204076Spjd } 1110204076Spjd 1111211977Spjd hook_init(); 1112211977Spjd 1113204076Spjd main_loop(); 1114204076Spjd 1115204076Spjd exit(0); 1116204076Spjd} 1117