hastd.c revision 219900
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 219900 2011-03-23 11:09:04Z 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"); 135219864Spjd else if (S_ISCHR(mode)) 136218044Spjd return ("character device"); 137219864Spjd else if (S_ISDIR(mode)) 138218044Spjd return ("directory"); 139218044Spjd else if (S_ISFIFO(mode)) 140218044Spjd return ("pipe or FIFO"); 141219864Spjd else if (S_ISLNK(mode)) 142218044Spjd return ("symbolic link"); 143219864Spjd else if (S_ISREG(mode)) 144218044Spjd return ("regular file"); 145218044Spjd else if (S_ISSOCK(mode)) 146218044Spjd return ("socket"); 147219864Spjd 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 } 227219900Spjd } else if (res->hr_role == HAST_ROLE_PRIMARY && 228219900Spjd fd == proto_descriptor(res->hr_conn)) { 229218218Spjd if (!isopen) { 230218375Spjd (void)snprintf(msg, sizeof(msg), 231218218Spjd "Descriptor %d (conn) is closed, but should be open.", 232218218Spjd fd); 233218218Spjd break; 234218218Spjd } 235218218Spjd if (!S_ISSOCK(mode)) { 236218375Spjd (void)snprintf(msg, sizeof(msg), 237218218Spjd "Descriptor %d (conn) is %s, but should be %s.", 238218218Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 239218218Spjd break; 240218218Spjd } 241218044Spjd } else if (res->hr_role == HAST_ROLE_SECONDARY && 242219900Spjd res->hr_conn != NULL && 243219900Spjd fd == proto_descriptor(res->hr_conn)) { 244219900Spjd if (isopen) { 245219900Spjd (void)snprintf(msg, sizeof(msg), 246219900Spjd "Descriptor %d (conn) is open, but should be closed.", 247219900Spjd fd); 248219900Spjd break; 249219900Spjd } 250219900Spjd } else if (res->hr_role == HAST_ROLE_SECONDARY && 251218044Spjd fd == proto_descriptor(res->hr_remotein)) { 252218044Spjd if (!isopen) { 253218375Spjd (void)snprintf(msg, sizeof(msg), 254218044Spjd "Descriptor %d (remote in) is closed, but should be open.", 255218044Spjd fd); 256218044Spjd break; 257218044Spjd } 258218044Spjd if (!S_ISSOCK(mode)) { 259218375Spjd (void)snprintf(msg, sizeof(msg), 260218044Spjd "Descriptor %d (remote in) is %s, but should be %s.", 261218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 262218044Spjd break; 263218044Spjd } 264218044Spjd } else if (res->hr_role == HAST_ROLE_SECONDARY && 265218044Spjd fd == proto_descriptor(res->hr_remoteout)) { 266218044Spjd if (!isopen) { 267218375Spjd (void)snprintf(msg, sizeof(msg), 268218044Spjd "Descriptor %d (remote out) is closed, but should be open.", 269218044Spjd fd); 270218044Spjd break; 271218044Spjd } 272218044Spjd if (!S_ISSOCK(mode)) { 273218375Spjd (void)snprintf(msg, sizeof(msg), 274218044Spjd "Descriptor %d (remote out) is %s, but should be %s.", 275218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 276218044Spjd break; 277218044Spjd } 278218044Spjd } else { 279218044Spjd if (isopen) { 280218375Spjd (void)snprintf(msg, sizeof(msg), 281218044Spjd "Descriptor %d is open (%s), but should be closed.", 282218044Spjd fd, dtype2str(mode)); 283218044Spjd break; 284218044Spjd } 285218044Spjd } 286218044Spjd } 287218044Spjd if (msg[0] != '\0') { 288218044Spjd pjdlog_init(pjdlogmode); 289218044Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 290218044Spjd role2str(res->hr_role)); 291218044Spjd PJDLOG_ABORT("%s", msg); 292218044Spjd } 293218044Spjd} 294218044Spjd 295204076Spjdstatic void 296207372Spjdchild_exit_log(unsigned int pid, int status) 297207372Spjd{ 298207372Spjd 299207372Spjd if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 300207372Spjd pjdlog_debug(1, "Worker process exited gracefully (pid=%u).", 301207372Spjd pid); 302207372Spjd } else if (WIFSIGNALED(status)) { 303207372Spjd pjdlog_error("Worker process killed (pid=%u, signal=%d).", 304207372Spjd pid, WTERMSIG(status)); 305207372Spjd } else { 306207372Spjd pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).", 307207372Spjd pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1); 308207372Spjd } 309207372Spjd} 310207372Spjd 311207372Spjdstatic void 312204076Spjdchild_exit(void) 313204076Spjd{ 314204076Spjd struct hast_resource *res; 315204076Spjd int status; 316204076Spjd pid_t pid; 317204076Spjd 318204076Spjd while ((pid = wait3(&status, WNOHANG, NULL)) > 0) { 319204076Spjd /* Find resource related to the process that just exited. */ 320204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 321204076Spjd if (pid == res->hr_workerpid) 322204076Spjd break; 323204076Spjd } 324204076Spjd if (res == NULL) { 325204076Spjd /* 326204076Spjd * This can happen when new connection arrives and we 327211977Spjd * cancel child responsible for the old one or if this 328211977Spjd * was hook which we executed. 329204076Spjd */ 330211977Spjd hook_check_one(pid, status); 331204076Spjd continue; 332204076Spjd } 333204076Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 334204076Spjd role2str(res->hr_role)); 335207372Spjd child_exit_log(pid, status); 336213006Spjd child_cleanup(res); 337204076Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 338207372Spjd /* 339207372Spjd * Restart child process if it was killed by signal 340207372Spjd * or exited because of temporary problem. 341207372Spjd */ 342207372Spjd if (WIFSIGNALED(status) || 343207372Spjd (WIFEXITED(status) && 344207372Spjd WEXITSTATUS(status) == EX_TEMPFAIL)) { 345207348Spjd sleep(1); 346207348Spjd pjdlog_info("Restarting worker process."); 347207348Spjd hastd_primary(res); 348207348Spjd } else { 349207348Spjd res->hr_role = HAST_ROLE_INIT; 350207348Spjd pjdlog_info("Changing resource role back to %s.", 351207348Spjd role2str(res->hr_role)); 352207348Spjd } 353204076Spjd } 354204076Spjd pjdlog_prefix_set("%s", ""); 355204076Spjd } 356204076Spjd} 357204076Spjd 358210886Spjdstatic bool 359210886Spjdresource_needs_restart(const struct hast_resource *res0, 360210886Spjd const struct hast_resource *res1) 361210886Spjd{ 362210886Spjd 363218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 364210886Spjd 365210886Spjd if (strcmp(res0->hr_provname, res1->hr_provname) != 0) 366210886Spjd return (true); 367210886Spjd if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0) 368210886Spjd return (true); 369210886Spjd if (res0->hr_role == HAST_ROLE_INIT || 370210886Spjd res0->hr_role == HAST_ROLE_SECONDARY) { 371210886Spjd if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 372210886Spjd return (true); 373219818Spjd if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0) 374219818Spjd return (true); 375210886Spjd if (res0->hr_replication != res1->hr_replication) 376210886Spjd return (true); 377219351Spjd if (res0->hr_checksum != res1->hr_checksum) 378219351Spjd return (true); 379219354Spjd if (res0->hr_compression != res1->hr_compression) 380219354Spjd return (true); 381210886Spjd if (res0->hr_timeout != res1->hr_timeout) 382210886Spjd return (true); 383211886Spjd if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 384211886Spjd return (true); 385210886Spjd } 386210886Spjd return (false); 387210886Spjd} 388210886Spjd 389210886Spjdstatic bool 390210886Spjdresource_needs_reload(const struct hast_resource *res0, 391210886Spjd const struct hast_resource *res1) 392210886Spjd{ 393210886Spjd 394218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 395218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0); 396218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0); 397210886Spjd 398210886Spjd if (res0->hr_role != HAST_ROLE_PRIMARY) 399210886Spjd return (false); 400210886Spjd 401210886Spjd if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 402210886Spjd return (true); 403219818Spjd if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0) 404219818Spjd return (true); 405210886Spjd if (res0->hr_replication != res1->hr_replication) 406210886Spjd return (true); 407219351Spjd if (res0->hr_checksum != res1->hr_checksum) 408219351Spjd return (true); 409219354Spjd if (res0->hr_compression != res1->hr_compression) 410219354Spjd return (true); 411210886Spjd if (res0->hr_timeout != res1->hr_timeout) 412210886Spjd return (true); 413211886Spjd if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 414211886Spjd return (true); 415210886Spjd return (false); 416210886Spjd} 417210886Spjd 418204076Spjdstatic void 419217784Spjdresource_reload(const struct hast_resource *res) 420217784Spjd{ 421217784Spjd struct nv *nvin, *nvout; 422217784Spjd int error; 423217784Spjd 424218138Spjd PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 425217784Spjd 426217784Spjd nvout = nv_alloc(); 427217784Spjd nv_add_uint8(nvout, HASTCTL_RELOAD, "cmd"); 428217784Spjd nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr"); 429219818Spjd nv_add_string(nvout, res->hr_sourceaddr, "sourceaddr"); 430217784Spjd nv_add_int32(nvout, (int32_t)res->hr_replication, "replication"); 431219351Spjd nv_add_int32(nvout, (int32_t)res->hr_checksum, "checksum"); 432219354Spjd nv_add_int32(nvout, (int32_t)res->hr_compression, "compression"); 433217784Spjd nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout"); 434217784Spjd nv_add_string(nvout, res->hr_exec, "exec"); 435217784Spjd if (nv_error(nvout) != 0) { 436217784Spjd nv_free(nvout); 437217784Spjd pjdlog_error("Unable to allocate header for reload message."); 438217784Spjd return; 439217784Spjd } 440217784Spjd if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) { 441217784Spjd pjdlog_errno(LOG_ERR, "Unable to send reload message"); 442217784Spjd nv_free(nvout); 443217784Spjd return; 444217784Spjd } 445217784Spjd nv_free(nvout); 446217784Spjd 447217784Spjd /* Receive response. */ 448217784Spjd if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) { 449217784Spjd pjdlog_errno(LOG_ERR, "Unable to receive reload reply"); 450217784Spjd return; 451217784Spjd } 452217784Spjd error = nv_get_int16(nvin, "error"); 453217784Spjd nv_free(nvin); 454217784Spjd if (error != 0) { 455217784Spjd pjdlog_common(LOG_ERR, 0, error, "Reload failed"); 456217784Spjd return; 457217784Spjd } 458217784Spjd} 459217784Spjd 460217784Spjdstatic void 461204076Spjdhastd_reload(void) 462204076Spjd{ 463210886Spjd struct hastd_config *newcfg; 464210886Spjd struct hast_resource *nres, *cres, *tres; 465210886Spjd uint8_t role; 466204076Spjd 467210886Spjd pjdlog_info("Reloading configuration..."); 468210886Spjd 469210886Spjd newcfg = yy_config_parse(cfgpath, false); 470210886Spjd if (newcfg == NULL) 471210886Spjd goto failed; 472210886Spjd 473210886Spjd /* 474210886Spjd * Check if control address has changed. 475210886Spjd */ 476210886Spjd if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 477210886Spjd if (proto_server(newcfg->hc_controladdr, 478210886Spjd &newcfg->hc_controlconn) < 0) { 479210886Spjd pjdlog_errno(LOG_ERR, 480210886Spjd "Unable to listen on control address %s", 481210886Spjd newcfg->hc_controladdr); 482210886Spjd goto failed; 483210886Spjd } 484210886Spjd } 485210886Spjd /* 486210886Spjd * Check if listen address has changed. 487210886Spjd */ 488210886Spjd if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) { 489210886Spjd if (proto_server(newcfg->hc_listenaddr, 490210886Spjd &newcfg->hc_listenconn) < 0) { 491210886Spjd pjdlog_errno(LOG_ERR, "Unable to listen on address %s", 492210886Spjd newcfg->hc_listenaddr); 493210886Spjd goto failed; 494210886Spjd } 495210886Spjd } 496210886Spjd /* 497210886Spjd * Only when both control and listen sockets are successfully 498210886Spjd * initialized switch them to new configuration. 499210886Spjd */ 500210886Spjd if (newcfg->hc_controlconn != NULL) { 501210886Spjd pjdlog_info("Control socket changed from %s to %s.", 502210886Spjd cfg->hc_controladdr, newcfg->hc_controladdr); 503210886Spjd proto_close(cfg->hc_controlconn); 504210886Spjd cfg->hc_controlconn = newcfg->hc_controlconn; 505210886Spjd newcfg->hc_controlconn = NULL; 506210886Spjd strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 507210886Spjd sizeof(cfg->hc_controladdr)); 508210886Spjd } 509210886Spjd if (newcfg->hc_listenconn != NULL) { 510210886Spjd pjdlog_info("Listen socket changed from %s to %s.", 511210886Spjd cfg->hc_listenaddr, newcfg->hc_listenaddr); 512210886Spjd proto_close(cfg->hc_listenconn); 513210886Spjd cfg->hc_listenconn = newcfg->hc_listenconn; 514210886Spjd newcfg->hc_listenconn = NULL; 515210886Spjd strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr, 516210886Spjd sizeof(cfg->hc_listenaddr)); 517210886Spjd } 518210886Spjd 519210886Spjd /* 520210886Spjd * Stop and remove resources that were removed from the configuration. 521210886Spjd */ 522210886Spjd TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 523210886Spjd TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 524210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 525210886Spjd break; 526210886Spjd } 527210886Spjd if (nres == NULL) { 528210886Spjd control_set_role(cres, HAST_ROLE_INIT); 529210886Spjd TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 530210886Spjd pjdlog_info("Resource %s removed.", cres->hr_name); 531210886Spjd free(cres); 532210886Spjd } 533210886Spjd } 534210886Spjd /* 535210886Spjd * Move new resources to the current configuration. 536210886Spjd */ 537210886Spjd TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 538210886Spjd TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 539210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 540210886Spjd break; 541210886Spjd } 542210886Spjd if (cres == NULL) { 543210886Spjd TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 544210886Spjd TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 545210886Spjd pjdlog_info("Resource %s added.", nres->hr_name); 546210886Spjd } 547210886Spjd } 548210886Spjd /* 549210886Spjd * Deal with modified resources. 550210886Spjd * Depending on what has changed exactly we might want to perform 551210886Spjd * different actions. 552210886Spjd * 553210886Spjd * We do full resource restart in the following situations: 554210886Spjd * Resource role is INIT or SECONDARY. 555210886Spjd * Resource role is PRIMARY and path to local component or provider 556210886Spjd * name has changed. 557210886Spjd * In case of PRIMARY, the worker process will be killed and restarted, 558210886Spjd * which also means removing /dev/hast/<name> provider and 559210886Spjd * recreating it. 560210886Spjd * 561210886Spjd * We do just reload (send SIGHUP to worker process) if we act as 562217729Spjd * PRIMARY, but only if remote address, replication mode, timeout or 563217729Spjd * execution path has changed. For those, there is no need to restart 564217729Spjd * worker process. 565210886Spjd * If PRIMARY receives SIGHUP, it will reconnect if remote address or 566210886Spjd * replication mode has changed or simply set new timeout if only 567210886Spjd * timeout has changed. 568210886Spjd */ 569210886Spjd TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 570210886Spjd TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 571210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 572210886Spjd break; 573210886Spjd } 574218138Spjd PJDLOG_ASSERT(cres != NULL); 575210886Spjd if (resource_needs_restart(cres, nres)) { 576210886Spjd pjdlog_info("Resource %s configuration was modified, restarting it.", 577210886Spjd cres->hr_name); 578210886Spjd role = cres->hr_role; 579210886Spjd control_set_role(cres, HAST_ROLE_INIT); 580210886Spjd TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 581210886Spjd free(cres); 582210886Spjd TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 583210886Spjd TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 584210886Spjd control_set_role(nres, role); 585210886Spjd } else if (resource_needs_reload(cres, nres)) { 586210886Spjd pjdlog_info("Resource %s configuration was modified, reloading it.", 587210886Spjd cres->hr_name); 588210886Spjd strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 589210886Spjd sizeof(cres->hr_remoteaddr)); 590219818Spjd strlcpy(cres->hr_sourceaddr, nres->hr_sourceaddr, 591219818Spjd sizeof(cres->hr_sourceaddr)); 592210886Spjd cres->hr_replication = nres->hr_replication; 593219351Spjd cres->hr_checksum = nres->hr_checksum; 594219354Spjd cres->hr_compression = nres->hr_compression; 595210886Spjd cres->hr_timeout = nres->hr_timeout; 596217729Spjd strlcpy(cres->hr_exec, nres->hr_exec, 597217729Spjd sizeof(cres->hr_exec)); 598217784Spjd if (cres->hr_workerpid != 0) 599217784Spjd resource_reload(cres); 600210886Spjd } 601210886Spjd } 602210886Spjd 603210886Spjd yy_config_free(newcfg); 604210886Spjd pjdlog_info("Configuration reloaded successfully."); 605210886Spjd return; 606210886Spjdfailed: 607210886Spjd if (newcfg != NULL) { 608210886Spjd if (newcfg->hc_controlconn != NULL) 609210886Spjd proto_close(newcfg->hc_controlconn); 610210886Spjd if (newcfg->hc_listenconn != NULL) 611210886Spjd proto_close(newcfg->hc_listenconn); 612210886Spjd yy_config_free(newcfg); 613210886Spjd } 614210886Spjd pjdlog_warning("Configuration not reloaded."); 615204076Spjd} 616204076Spjd 617204076Spjdstatic void 618211899Spjdterminate_workers(void) 619211899Spjd{ 620211899Spjd struct hast_resource *res; 621211899Spjd 622211899Spjd pjdlog_info("Termination signal received, exiting."); 623211899Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 624211899Spjd if (res->hr_workerpid == 0) 625211899Spjd continue; 626211899Spjd pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 627211899Spjd res->hr_name, role2str(res->hr_role), res->hr_workerpid); 628211899Spjd if (kill(res->hr_workerpid, SIGTERM) == 0) 629211899Spjd continue; 630211899Spjd pjdlog_errno(LOG_WARNING, 631211899Spjd "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 632211899Spjd res->hr_name, role2str(res->hr_role), res->hr_workerpid); 633211899Spjd } 634211899Spjd} 635211899Spjd 636211899Spjdstatic void 637204076Spjdlisten_accept(void) 638204076Spjd{ 639204076Spjd struct hast_resource *res; 640204076Spjd struct proto_conn *conn; 641204076Spjd struct nv *nvin, *nvout, *nverr; 642204076Spjd const char *resname; 643204076Spjd const unsigned char *token; 644204076Spjd char laddr[256], raddr[256]; 645204076Spjd size_t size; 646204076Spjd pid_t pid; 647204076Spjd int status; 648204076Spjd 649204076Spjd proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr)); 650204076Spjd pjdlog_debug(1, "Accepting connection to %s.", laddr); 651204076Spjd 652204076Spjd if (proto_accept(cfg->hc_listenconn, &conn) < 0) { 653204076Spjd pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 654204076Spjd return; 655204076Spjd } 656204076Spjd 657204076Spjd proto_local_address(conn, laddr, sizeof(laddr)); 658204076Spjd proto_remote_address(conn, raddr, sizeof(raddr)); 659209185Spjd pjdlog_info("Connection from %s to %s.", raddr, laddr); 660204076Spjd 661207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 662207371Spjd if (proto_timeout(conn, HAST_TIMEOUT) < 0) 663207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 664207371Spjd 665204076Spjd nvin = nvout = nverr = NULL; 666204076Spjd 667204076Spjd /* 668204076Spjd * Before receiving any data see if remote host have access to any 669204076Spjd * resource. 670204076Spjd */ 671204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 672204076Spjd if (proto_address_match(conn, res->hr_remoteaddr)) 673204076Spjd break; 674204076Spjd } 675204076Spjd if (res == NULL) { 676204076Spjd pjdlog_error("Client %s isn't known.", raddr); 677204076Spjd goto close; 678204076Spjd } 679204076Spjd /* Ok, remote host can access at least one resource. */ 680204076Spjd 681204076Spjd if (hast_proto_recv_hdr(conn, &nvin) < 0) { 682204076Spjd pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 683204076Spjd raddr); 684204076Spjd goto close; 685204076Spjd } 686204076Spjd 687204076Spjd resname = nv_get_string(nvin, "resource"); 688204076Spjd if (resname == NULL) { 689204076Spjd pjdlog_error("No 'resource' field in the header received from %s.", 690204076Spjd raddr); 691204076Spjd goto close; 692204076Spjd } 693204076Spjd pjdlog_debug(2, "%s: resource=%s", raddr, resname); 694204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 695204076Spjd /* 696204076Spjd * NULL token means that this is first conection. 697204076Spjd */ 698204076Spjd if (token != NULL && size != sizeof(res->hr_token)) { 699204076Spjd pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 700204076Spjd raddr, sizeof(res->hr_token), size); 701204076Spjd goto close; 702204076Spjd } 703204076Spjd 704204076Spjd /* 705204076Spjd * From now on we want to send errors to the remote node. 706204076Spjd */ 707204076Spjd nverr = nv_alloc(); 708204076Spjd 709204076Spjd /* Find resource related to this connection. */ 710204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 711204076Spjd if (strcmp(resname, res->hr_name) == 0) 712204076Spjd break; 713204076Spjd } 714204076Spjd /* Have we found the resource? */ 715204076Spjd if (res == NULL) { 716204076Spjd pjdlog_error("No resource '%s' as requested by %s.", 717204076Spjd resname, raddr); 718204076Spjd nv_add_stringf(nverr, "errmsg", "Resource not configured."); 719204076Spjd goto fail; 720204076Spjd } 721204076Spjd 722204076Spjd /* Now that we know resource name setup log prefix. */ 723204076Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 724204076Spjd 725204076Spjd /* Does the remote host have access to this resource? */ 726204076Spjd if (!proto_address_match(conn, res->hr_remoteaddr)) { 727204076Spjd pjdlog_error("Client %s has no access to the resource.", raddr); 728204076Spjd nv_add_stringf(nverr, "errmsg", "No access to the resource."); 729204076Spjd goto fail; 730204076Spjd } 731204076Spjd /* Is the resource marked as secondary? */ 732204076Spjd if (res->hr_role != HAST_ROLE_SECONDARY) { 733204076Spjd pjdlog_error("We act as %s for the resource and not as %s as requested by %s.", 734204076Spjd role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 735204076Spjd raddr); 736204076Spjd nv_add_stringf(nverr, "errmsg", 737204076Spjd "Remote node acts as %s for the resource and not as %s.", 738204076Spjd role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 739204076Spjd goto fail; 740204076Spjd } 741204076Spjd /* Does token (if exists) match? */ 742204076Spjd if (token != NULL && memcmp(token, res->hr_token, 743204076Spjd sizeof(res->hr_token)) != 0) { 744204076Spjd pjdlog_error("Token received from %s doesn't match.", raddr); 745209185Spjd nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 746204076Spjd goto fail; 747204076Spjd } 748204076Spjd /* 749204076Spjd * If there is no token, but we have half-open connection 750204076Spjd * (only remotein) or full connection (worker process is running) 751204076Spjd * we have to cancel those and accept the new connection. 752204076Spjd */ 753204076Spjd if (token == NULL) { 754218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 755204076Spjd pjdlog_debug(1, "Initial connection from %s.", raddr); 756204076Spjd if (res->hr_workerpid != 0) { 757218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 758204076Spjd pjdlog_debug(1, 759204076Spjd "Worker process exists (pid=%u), stopping it.", 760204076Spjd (unsigned int)res->hr_workerpid); 761204076Spjd /* Stop child process. */ 762204076Spjd if (kill(res->hr_workerpid, SIGINT) < 0) { 763204076Spjd pjdlog_errno(LOG_ERR, 764204076Spjd "Unable to stop worker process (pid=%u)", 765204076Spjd (unsigned int)res->hr_workerpid); 766204076Spjd /* 767204076Spjd * Other than logging the problem we 768204076Spjd * ignore it - nothing smart to do. 769204076Spjd */ 770204076Spjd } 771204076Spjd /* Wait for it to exit. */ 772204076Spjd else if ((pid = waitpid(res->hr_workerpid, 773204076Spjd &status, 0)) != res->hr_workerpid) { 774207372Spjd /* We can only log the problem. */ 775204076Spjd pjdlog_errno(LOG_ERR, 776204076Spjd "Waiting for worker process (pid=%u) failed", 777204076Spjd (unsigned int)res->hr_workerpid); 778204076Spjd } else { 779207372Spjd child_exit_log(res->hr_workerpid, status); 780204076Spjd } 781213006Spjd child_cleanup(res); 782204076Spjd } else if (res->hr_remotein != NULL) { 783204076Spjd char oaddr[256]; 784204076Spjd 785213981Spjd proto_remote_address(res->hr_remotein, oaddr, 786213981Spjd sizeof(oaddr)); 787204076Spjd pjdlog_debug(1, 788204076Spjd "Canceling half-open connection from %s on connection from %s.", 789204076Spjd oaddr, raddr); 790204076Spjd proto_close(res->hr_remotein); 791204076Spjd res->hr_remotein = NULL; 792204076Spjd } 793204076Spjd } 794204076Spjd 795204076Spjd /* 796204076Spjd * Checks and cleanups are done. 797204076Spjd */ 798204076Spjd 799204076Spjd if (token == NULL) { 800204076Spjd arc4random_buf(res->hr_token, sizeof(res->hr_token)); 801204076Spjd nvout = nv_alloc(); 802204076Spjd nv_add_uint8_array(nvout, res->hr_token, 803204076Spjd sizeof(res->hr_token), "token"); 804204076Spjd if (nv_error(nvout) != 0) { 805204076Spjd pjdlog_common(LOG_ERR, 0, nv_error(nvout), 806204076Spjd "Unable to prepare return header for %s", raddr); 807204076Spjd nv_add_stringf(nverr, "errmsg", 808204076Spjd "Remote node was unable to prepare return header: %s.", 809204076Spjd strerror(nv_error(nvout))); 810204076Spjd goto fail; 811204076Spjd } 812204076Spjd if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) { 813204076Spjd int error = errno; 814204076Spjd 815204076Spjd pjdlog_errno(LOG_ERR, "Unable to send response to %s", 816204076Spjd raddr); 817204076Spjd nv_add_stringf(nverr, "errmsg", 818204076Spjd "Remote node was unable to send response: %s.", 819204076Spjd strerror(error)); 820204076Spjd goto fail; 821204076Spjd } 822204076Spjd res->hr_remotein = conn; 823204076Spjd pjdlog_debug(1, "Incoming connection from %s configured.", 824204076Spjd raddr); 825204076Spjd } else { 826204076Spjd res->hr_remoteout = conn; 827204076Spjd pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 828204076Spjd hastd_secondary(res, nvin); 829204076Spjd } 830204076Spjd nv_free(nvin); 831204076Spjd nv_free(nvout); 832204076Spjd nv_free(nverr); 833204076Spjd pjdlog_prefix_set("%s", ""); 834204076Spjd return; 835204076Spjdfail: 836204076Spjd if (nv_error(nverr) != 0) { 837204076Spjd pjdlog_common(LOG_ERR, 0, nv_error(nverr), 838204076Spjd "Unable to prepare error header for %s", raddr); 839204076Spjd goto close; 840204076Spjd } 841204076Spjd if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) { 842204076Spjd pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 843204076Spjd goto close; 844204076Spjd } 845204076Spjdclose: 846204076Spjd if (nvin != NULL) 847204076Spjd nv_free(nvin); 848204076Spjd if (nvout != NULL) 849204076Spjd nv_free(nvout); 850204076Spjd if (nverr != NULL) 851204076Spjd nv_free(nverr); 852204076Spjd proto_close(conn); 853204076Spjd pjdlog_prefix_set("%s", ""); 854204076Spjd} 855204076Spjd 856204076Spjdstatic void 857218218Spjdconnection_migrate(struct hast_resource *res) 858218218Spjd{ 859218218Spjd struct proto_conn *conn; 860218218Spjd int16_t val = 0; 861218218Spjd 862219814Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 863219814Spjd 864219900Spjd PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 865219900Spjd 866218218Spjd if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) { 867218218Spjd pjdlog_errno(LOG_WARNING, 868218218Spjd "Unable to receive connection command"); 869218218Spjd return; 870218218Spjd } 871219818Spjd if (proto_client(res->hr_sourceaddr[0] != '\0' ? res->hr_sourceaddr : NULL, 872219818Spjd res->hr_remoteaddr, &conn) < 0) { 873218218Spjd val = errno; 874218218Spjd pjdlog_errno(LOG_WARNING, 875218218Spjd "Unable to create outgoing connection to %s", 876218218Spjd res->hr_remoteaddr); 877218218Spjd goto out; 878218218Spjd } 879218218Spjd if (proto_connect(conn, -1) < 0) { 880218218Spjd val = errno; 881218218Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 882218218Spjd res->hr_remoteaddr); 883218218Spjd proto_close(conn); 884218218Spjd goto out; 885218218Spjd } 886218218Spjd val = 0; 887218218Spjdout: 888218218Spjd if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) { 889218218Spjd pjdlog_errno(LOG_WARNING, 890218218Spjd "Unable to send reply to connection request"); 891218218Spjd } 892218218Spjd if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0) 893218218Spjd pjdlog_errno(LOG_WARNING, "Unable to send connection"); 894219814Spjd 895219814Spjd pjdlog_prefix_set("%s", ""); 896218218Spjd} 897218218Spjd 898218218Spjdstatic void 899219837Spjdcheck_signals(void) 900204076Spjd{ 901213009Spjd struct timespec sigtimeout; 902213009Spjd sigset_t mask; 903219837Spjd int signo; 904204076Spjd 905213009Spjd sigtimeout.tv_sec = 0; 906213009Spjd sigtimeout.tv_nsec = 0; 907211977Spjd 908213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 909213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 910213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 911213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 912213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 913213009Spjd 914219837Spjd while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 915219837Spjd switch (signo) { 916219837Spjd case SIGINT: 917219837Spjd case SIGTERM: 918219837Spjd sigexit_received = true; 919219837Spjd terminate_workers(); 920219837Spjd proto_close(cfg->hc_controlconn); 921219837Spjd exit(EX_OK); 922219837Spjd break; 923219837Spjd case SIGCHLD: 924219837Spjd child_exit(); 925219837Spjd break; 926219837Spjd case SIGHUP: 927219837Spjd hastd_reload(); 928219837Spjd break; 929219837Spjd default: 930219837Spjd PJDLOG_ABORT("Unexpected signal (%d).", signo); 931219837Spjd } 932219837Spjd } 933219837Spjd} 934219837Spjd 935219837Spjdstatic void 936219837Spjdmain_loop(void) 937219837Spjd{ 938219837Spjd struct hast_resource *res; 939219837Spjd struct timeval seltimeout; 940219837Spjd int fd, maxfd, ret; 941219837Spjd time_t lastcheck, now; 942219837Spjd fd_set rfds; 943219837Spjd 944219864Spjd lastcheck = time(NULL); 945219837Spjd seltimeout.tv_sec = REPORT_INTERVAL; 946219837Spjd seltimeout.tv_usec = 0; 947219837Spjd 948216477Spjd pjdlog_info("Started successfully, running protocol version %d.", 949216477Spjd HAST_PROTO_VERSION); 950216477Spjd 951204076Spjd for (;;) { 952219837Spjd check_signals(); 953204076Spjd 954209177Spjd /* Setup descriptors for select(2). */ 955204076Spjd FD_ZERO(&rfds); 956212038Spjd maxfd = fd = proto_descriptor(cfg->hc_controlconn); 957218138Spjd PJDLOG_ASSERT(fd >= 0); 958212038Spjd FD_SET(fd, &rfds); 959212038Spjd fd = proto_descriptor(cfg->hc_listenconn); 960218138Spjd PJDLOG_ASSERT(fd >= 0); 961212038Spjd FD_SET(fd, &rfds); 962212038Spjd maxfd = fd > maxfd ? fd : maxfd; 963212038Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 964212038Spjd if (res->hr_event == NULL) 965212038Spjd continue; 966212038Spjd fd = proto_descriptor(res->hr_event); 967218138Spjd PJDLOG_ASSERT(fd >= 0); 968212038Spjd FD_SET(fd, &rfds); 969212038Spjd maxfd = fd > maxfd ? fd : maxfd; 970218218Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 971218218Spjd /* Only primary workers asks for connections. */ 972219900Spjd PJDLOG_ASSERT(res->hr_conn != NULL); 973218218Spjd fd = proto_descriptor(res->hr_conn); 974218218Spjd PJDLOG_ASSERT(fd >= 0); 975218218Spjd FD_SET(fd, &rfds); 976218218Spjd maxfd = fd > maxfd ? fd : maxfd; 977219900Spjd } else { 978219900Spjd PJDLOG_ASSERT(res->hr_conn == NULL); 979218218Spjd } 980212038Spjd } 981204076Spjd 982218138Spjd PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE); 983213009Spjd ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 984219813Spjd now = time(NULL); 985219813Spjd if (lastcheck + REPORT_INTERVAL <= now) { 986213429Spjd hook_check(); 987219813Spjd lastcheck = now; 988219813Spjd } 989219813Spjd if (ret == 0) { 990219813Spjd /* 991219813Spjd * select(2) timed out, so there should be no 992219813Spjd * descriptors to check. 993219813Spjd */ 994219813Spjd continue; 995219813Spjd } else if (ret == -1) { 996204076Spjd if (errno == EINTR) 997204076Spjd continue; 998204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 999204076Spjd pjdlog_exit(EX_OSERR, "select() failed"); 1000204076Spjd } 1001204076Spjd 1002219837Spjd /* 1003219837Spjd * Check for signals before we do anything to update our 1004219837Spjd * info about terminated workers in the meantime. 1005219837Spjd */ 1006219837Spjd check_signals(); 1007219837Spjd 1008212038Spjd if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 1009204076Spjd control_handle(cfg); 1010212038Spjd if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds)) 1011204076Spjd listen_accept(); 1012212038Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 1013212038Spjd if (res->hr_event == NULL) 1014212038Spjd continue; 1015212038Spjd if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 1016212038Spjd if (event_recv(res) == 0) 1017212038Spjd continue; 1018212038Spjd /* The worker process exited? */ 1019212038Spjd proto_close(res->hr_event); 1020212038Spjd res->hr_event = NULL; 1021219900Spjd if (res->hr_conn != NULL) { 1022219900Spjd proto_close(res->hr_conn); 1023219900Spjd res->hr_conn = NULL; 1024219900Spjd } 1025218218Spjd continue; 1026212038Spjd } 1027219900Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 1028219900Spjd PJDLOG_ASSERT(res->hr_conn != NULL); 1029219900Spjd if (FD_ISSET(proto_descriptor(res->hr_conn), 1030219900Spjd &rfds)) { 1031219900Spjd connection_migrate(res); 1032219900Spjd } 1033219900Spjd } else { 1034219900Spjd PJDLOG_ASSERT(res->hr_conn == NULL); 1035218218Spjd } 1036212038Spjd } 1037204076Spjd } 1038204076Spjd} 1039204076Spjd 1040213428Spjdstatic void 1041213428Spjddummy_sighandler(int sig __unused) 1042213428Spjd{ 1043213428Spjd /* Nothing to do. */ 1044213428Spjd} 1045213428Spjd 1046204076Spjdint 1047204076Spjdmain(int argc, char *argv[]) 1048204076Spjd{ 1049204076Spjd const char *pidfile; 1050204076Spjd pid_t otherpid; 1051204076Spjd bool foreground; 1052204076Spjd int debuglevel; 1053213009Spjd sigset_t mask; 1054204076Spjd 1055204076Spjd foreground = false; 1056204076Spjd debuglevel = 0; 1057204076Spjd pidfile = HASTD_PIDFILE; 1058204076Spjd 1059204076Spjd for (;;) { 1060204076Spjd int ch; 1061204076Spjd 1062204076Spjd ch = getopt(argc, argv, "c:dFhP:"); 1063204076Spjd if (ch == -1) 1064204076Spjd break; 1065204076Spjd switch (ch) { 1066204076Spjd case 'c': 1067204076Spjd cfgpath = optarg; 1068204076Spjd break; 1069204076Spjd case 'd': 1070204076Spjd debuglevel++; 1071204076Spjd break; 1072204076Spjd case 'F': 1073204076Spjd foreground = true; 1074204076Spjd break; 1075204076Spjd case 'P': 1076204076Spjd pidfile = optarg; 1077204076Spjd break; 1078204076Spjd case 'h': 1079204076Spjd default: 1080204076Spjd usage(); 1081204076Spjd } 1082204076Spjd } 1083204076Spjd argc -= optind; 1084204076Spjd argv += optind; 1085204076Spjd 1086217965Spjd pjdlog_init(PJDLOG_MODE_STD); 1087204076Spjd pjdlog_debug_set(debuglevel); 1088204076Spjd 1089214273Spjd g_gate_load(); 1090214273Spjd 1091204076Spjd pfh = pidfile_open(pidfile, 0600, &otherpid); 1092204076Spjd if (pfh == NULL) { 1093204076Spjd if (errno == EEXIST) { 1094204076Spjd pjdlog_exitx(EX_TEMPFAIL, 1095204076Spjd "Another hastd is already running, pid: %jd.", 1096204076Spjd (intmax_t)otherpid); 1097204076Spjd } 1098204076Spjd /* If we cannot create pidfile from other reasons, only warn. */ 1099210879Spjd pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile"); 1100204076Spjd } 1101204076Spjd 1102210883Spjd cfg = yy_config_parse(cfgpath, true); 1103218138Spjd PJDLOG_ASSERT(cfg != NULL); 1104204076Spjd 1105213428Spjd /* 1106217307Spjd * Restore default actions for interesting signals in case parent 1107217307Spjd * process (like init(8)) decided to ignore some of them (like SIGHUP). 1108217307Spjd */ 1109217307Spjd PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 1110217307Spjd PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 1111217307Spjd PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 1112217307Spjd /* 1113213428Spjd * Because SIGCHLD is ignored by default, setup dummy handler for it, 1114213428Spjd * so we can mask it. 1115213428Spjd */ 1116213428Spjd PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 1117217307Spjd 1118213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1119213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1120213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1121213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1122213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1123213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 1124204076Spjd 1125204076Spjd /* Listen on control address. */ 1126204076Spjd if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) { 1127204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1128204076Spjd pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 1129204076Spjd cfg->hc_controladdr); 1130204076Spjd } 1131204076Spjd /* Listen for remote connections. */ 1132204076Spjd if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) { 1133204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1134204076Spjd pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 1135204076Spjd cfg->hc_listenaddr); 1136204076Spjd } 1137204076Spjd 1138204076Spjd if (!foreground) { 1139204076Spjd if (daemon(0, 0) < 0) { 1140204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1141204076Spjd pjdlog_exit(EX_OSERR, "Unable to daemonize"); 1142204076Spjd } 1143204076Spjd 1144204076Spjd /* Start logging to syslog. */ 1145204076Spjd pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 1146204076Spjd 1147204076Spjd /* Write PID to a file. */ 1148204076Spjd if (pidfile_write(pfh) < 0) { 1149204076Spjd pjdlog_errno(LOG_WARNING, 1150204076Spjd "Unable to write PID to a file"); 1151204076Spjd } 1152204076Spjd } 1153204076Spjd 1154211977Spjd hook_init(); 1155211977Spjd 1156204076Spjd main_loop(); 1157204076Spjd 1158204076Spjd exit(0); 1159204076Spjd} 1160