hastd.c revision 222108
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 222108 2011-05-19 23:18:42Z 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; 101222108Spjd struct hastd_listen *lst; 102218041Spjd 103218041Spjd TAILQ_FOREACH(tres, &cfg->hc_resources, hr_next) { 104218041Spjd if (tres == res) { 105218041Spjd PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY || 106218041Spjd (res->hr_remotein == NULL && 107218041Spjd res->hr_remoteout == NULL)); 108218041Spjd continue; 109218041Spjd } 110218041Spjd if (tres->hr_remotein != NULL) 111218041Spjd proto_close(tres->hr_remotein); 112218041Spjd if (tres->hr_remoteout != NULL) 113218041Spjd proto_close(tres->hr_remoteout); 114218370Spjd if (tres->hr_ctrl != NULL) 115218370Spjd proto_close(tres->hr_ctrl); 116218370Spjd if (tres->hr_event != NULL) 117218370Spjd proto_close(tres->hr_event); 118218370Spjd if (tres->hr_conn != NULL) 119218370Spjd proto_close(tres->hr_conn); 120218041Spjd } 121218041Spjd if (cfg->hc_controlin != NULL) 122218041Spjd proto_close(cfg->hc_controlin); 123218041Spjd proto_close(cfg->hc_controlconn); 124222108Spjd TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) { 125222108Spjd if (lst->hl_conn != NULL) 126222108Spjd proto_close(lst->hl_conn); 127222108Spjd } 128218041Spjd (void)pidfile_close(pfh); 129218041Spjd hook_fini(); 130218041Spjd pjdlog_fini(); 131218041Spjd} 132218041Spjd 133218044Spjdstatic const char * 134218044Spjddtype2str(mode_t mode) 135218044Spjd{ 136218044Spjd 137218044Spjd if (S_ISBLK(mode)) 138218044Spjd return ("block device"); 139219864Spjd else if (S_ISCHR(mode)) 140218044Spjd return ("character device"); 141219864Spjd else if (S_ISDIR(mode)) 142218044Spjd return ("directory"); 143218044Spjd else if (S_ISFIFO(mode)) 144218044Spjd return ("pipe or FIFO"); 145219864Spjd else if (S_ISLNK(mode)) 146218044Spjd return ("symbolic link"); 147219864Spjd else if (S_ISREG(mode)) 148218044Spjd return ("regular file"); 149218044Spjd else if (S_ISSOCK(mode)) 150218044Spjd return ("socket"); 151219864Spjd else if (S_ISWHT(mode)) 152218044Spjd return ("whiteout"); 153218044Spjd else 154218044Spjd return ("unknown"); 155218044Spjd} 156218044Spjd 157218044Spjdvoid 158218044Spjddescriptors_assert(const struct hast_resource *res, int pjdlogmode) 159218044Spjd{ 160218044Spjd char msg[256]; 161218044Spjd struct stat sb; 162218044Spjd long maxfd; 163218044Spjd bool isopen; 164218044Spjd mode_t mode; 165218044Spjd int fd; 166218044Spjd 167218044Spjd /* 168218044Spjd * At this point descriptor to syslog socket is closed, so if we want 169218044Spjd * to log assertion message, we have to first store it in 'msg' local 170218044Spjd * buffer and then open syslog socket and log it. 171218044Spjd */ 172218044Spjd msg[0] = '\0'; 173218044Spjd 174218044Spjd maxfd = sysconf(_SC_OPEN_MAX); 175218044Spjd if (maxfd < 0) { 176218373Spjd pjdlog_init(pjdlogmode); 177218373Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 178218373Spjd role2str(res->hr_role)); 179218044Spjd pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed"); 180218373Spjd pjdlog_fini(); 181218044Spjd maxfd = 16384; 182218044Spjd } 183218044Spjd for (fd = 0; fd <= maxfd; fd++) { 184218044Spjd if (fstat(fd, &sb) == 0) { 185218044Spjd isopen = true; 186218044Spjd mode = sb.st_mode; 187218044Spjd } else if (errno == EBADF) { 188218044Spjd isopen = false; 189218044Spjd mode = 0; 190218044Spjd } else { 191218375Spjd (void)snprintf(msg, sizeof(msg), 192218044Spjd "Unable to fstat descriptor %d: %s", fd, 193218044Spjd strerror(errno)); 194218374Spjd break; 195218044Spjd } 196218044Spjd if (fd == STDIN_FILENO || fd == STDOUT_FILENO || 197218044Spjd fd == STDERR_FILENO) { 198218044Spjd if (!isopen) { 199218375Spjd (void)snprintf(msg, sizeof(msg), 200218044Spjd "Descriptor %d (%s) is closed, but should be open.", 201218044Spjd fd, (fd == STDIN_FILENO ? "stdin" : 202218044Spjd (fd == STDOUT_FILENO ? "stdout" : "stderr"))); 203218044Spjd break; 204218044Spjd } 205218044Spjd } else if (fd == proto_descriptor(res->hr_event)) { 206218044Spjd if (!isopen) { 207218375Spjd (void)snprintf(msg, sizeof(msg), 208218044Spjd "Descriptor %d (event) is closed, but should be open.", 209218044Spjd fd); 210218044Spjd break; 211218044Spjd } 212218044Spjd if (!S_ISSOCK(mode)) { 213218375Spjd (void)snprintf(msg, sizeof(msg), 214218044Spjd "Descriptor %d (event) is %s, but should be %s.", 215218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 216218044Spjd break; 217218044Spjd } 218218044Spjd } else if (fd == proto_descriptor(res->hr_ctrl)) { 219218044Spjd if (!isopen) { 220218375Spjd (void)snprintf(msg, sizeof(msg), 221218044Spjd "Descriptor %d (ctrl) is closed, but should be open.", 222218044Spjd fd); 223218044Spjd break; 224218044Spjd } 225218044Spjd if (!S_ISSOCK(mode)) { 226218375Spjd (void)snprintf(msg, sizeof(msg), 227218044Spjd "Descriptor %d (ctrl) is %s, but should be %s.", 228218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 229218044Spjd break; 230218044Spjd } 231219900Spjd } else if (res->hr_role == HAST_ROLE_PRIMARY && 232219900Spjd fd == proto_descriptor(res->hr_conn)) { 233218218Spjd if (!isopen) { 234218375Spjd (void)snprintf(msg, sizeof(msg), 235218218Spjd "Descriptor %d (conn) is closed, but should be open.", 236218218Spjd fd); 237218218Spjd break; 238218218Spjd } 239218218Spjd if (!S_ISSOCK(mode)) { 240218375Spjd (void)snprintf(msg, sizeof(msg), 241218218Spjd "Descriptor %d (conn) is %s, but should be %s.", 242218218Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 243218218Spjd break; 244218218Spjd } 245218044Spjd } else if (res->hr_role == HAST_ROLE_SECONDARY && 246219900Spjd res->hr_conn != NULL && 247219900Spjd fd == proto_descriptor(res->hr_conn)) { 248219900Spjd if (isopen) { 249219900Spjd (void)snprintf(msg, sizeof(msg), 250219900Spjd "Descriptor %d (conn) is open, but should be closed.", 251219900Spjd fd); 252219900Spjd break; 253219900Spjd } 254219900Spjd } else if (res->hr_role == HAST_ROLE_SECONDARY && 255218044Spjd fd == proto_descriptor(res->hr_remotein)) { 256218044Spjd if (!isopen) { 257218375Spjd (void)snprintf(msg, sizeof(msg), 258218044Spjd "Descriptor %d (remote in) 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 in) is %s, but should be %s.", 265218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 266218044Spjd break; 267218044Spjd } 268218044Spjd } else if (res->hr_role == HAST_ROLE_SECONDARY && 269218044Spjd fd == proto_descriptor(res->hr_remoteout)) { 270218044Spjd if (!isopen) { 271218375Spjd (void)snprintf(msg, sizeof(msg), 272218044Spjd "Descriptor %d (remote out) is closed, but should be open.", 273218044Spjd fd); 274218044Spjd break; 275218044Spjd } 276218044Spjd if (!S_ISSOCK(mode)) { 277218375Spjd (void)snprintf(msg, sizeof(msg), 278218044Spjd "Descriptor %d (remote out) is %s, but should be %s.", 279218044Spjd fd, dtype2str(mode), dtype2str(S_IFSOCK)); 280218044Spjd break; 281218044Spjd } 282218044Spjd } else { 283218044Spjd if (isopen) { 284218375Spjd (void)snprintf(msg, sizeof(msg), 285218044Spjd "Descriptor %d is open (%s), but should be closed.", 286218044Spjd fd, dtype2str(mode)); 287218044Spjd break; 288218044Spjd } 289218044Spjd } 290218044Spjd } 291218044Spjd if (msg[0] != '\0') { 292218044Spjd pjdlog_init(pjdlogmode); 293218044Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 294218044Spjd role2str(res->hr_role)); 295218044Spjd PJDLOG_ABORT("%s", msg); 296218044Spjd } 297218044Spjd} 298218044Spjd 299204076Spjdstatic void 300207372Spjdchild_exit_log(unsigned int pid, int status) 301207372Spjd{ 302207372Spjd 303207372Spjd if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 304207372Spjd pjdlog_debug(1, "Worker process exited gracefully (pid=%u).", 305207372Spjd pid); 306207372Spjd } else if (WIFSIGNALED(status)) { 307207372Spjd pjdlog_error("Worker process killed (pid=%u, signal=%d).", 308207372Spjd pid, WTERMSIG(status)); 309207372Spjd } else { 310207372Spjd pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).", 311207372Spjd pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1); 312207372Spjd } 313207372Spjd} 314207372Spjd 315207372Spjdstatic void 316204076Spjdchild_exit(void) 317204076Spjd{ 318204076Spjd struct hast_resource *res; 319204076Spjd int status; 320204076Spjd pid_t pid; 321204076Spjd 322204076Spjd while ((pid = wait3(&status, WNOHANG, NULL)) > 0) { 323204076Spjd /* Find resource related to the process that just exited. */ 324204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 325204076Spjd if (pid == res->hr_workerpid) 326204076Spjd break; 327204076Spjd } 328204076Spjd if (res == NULL) { 329204076Spjd /* 330204076Spjd * This can happen when new connection arrives and we 331211977Spjd * cancel child responsible for the old one or if this 332211977Spjd * was hook which we executed. 333204076Spjd */ 334211977Spjd hook_check_one(pid, status); 335204076Spjd continue; 336204076Spjd } 337204076Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 338204076Spjd role2str(res->hr_role)); 339207372Spjd child_exit_log(pid, status); 340213006Spjd child_cleanup(res); 341204076Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 342207372Spjd /* 343207372Spjd * Restart child process if it was killed by signal 344207372Spjd * or exited because of temporary problem. 345207372Spjd */ 346207372Spjd if (WIFSIGNALED(status) || 347207372Spjd (WIFEXITED(status) && 348207372Spjd WEXITSTATUS(status) == EX_TEMPFAIL)) { 349207348Spjd sleep(1); 350207348Spjd pjdlog_info("Restarting worker process."); 351207348Spjd hastd_primary(res); 352207348Spjd } else { 353207348Spjd res->hr_role = HAST_ROLE_INIT; 354207348Spjd pjdlog_info("Changing resource role back to %s.", 355207348Spjd role2str(res->hr_role)); 356207348Spjd } 357204076Spjd } 358204076Spjd pjdlog_prefix_set("%s", ""); 359204076Spjd } 360204076Spjd} 361204076Spjd 362210886Spjdstatic bool 363210886Spjdresource_needs_restart(const struct hast_resource *res0, 364210886Spjd const struct hast_resource *res1) 365210886Spjd{ 366210886Spjd 367218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 368210886Spjd 369210886Spjd if (strcmp(res0->hr_provname, res1->hr_provname) != 0) 370210886Spjd return (true); 371210886Spjd if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0) 372210886Spjd return (true); 373210886Spjd if (res0->hr_role == HAST_ROLE_INIT || 374210886Spjd res0->hr_role == HAST_ROLE_SECONDARY) { 375210886Spjd if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 376210886Spjd return (true); 377219818Spjd if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0) 378219818Spjd return (true); 379210886Spjd if (res0->hr_replication != res1->hr_replication) 380210886Spjd return (true); 381219351Spjd if (res0->hr_checksum != res1->hr_checksum) 382219351Spjd return (true); 383219354Spjd if (res0->hr_compression != res1->hr_compression) 384219354Spjd return (true); 385210886Spjd if (res0->hr_timeout != res1->hr_timeout) 386210886Spjd return (true); 387211886Spjd if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 388211886Spjd return (true); 389210886Spjd } 390210886Spjd return (false); 391210886Spjd} 392210886Spjd 393210886Spjdstatic bool 394210886Spjdresource_needs_reload(const struct hast_resource *res0, 395210886Spjd const struct hast_resource *res1) 396210886Spjd{ 397210886Spjd 398218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 399218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0); 400218138Spjd PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0); 401210886Spjd 402210886Spjd if (res0->hr_role != HAST_ROLE_PRIMARY) 403210886Spjd return (false); 404210886Spjd 405210886Spjd if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 406210886Spjd return (true); 407219818Spjd if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0) 408219818Spjd return (true); 409210886Spjd if (res0->hr_replication != res1->hr_replication) 410210886Spjd return (true); 411219351Spjd if (res0->hr_checksum != res1->hr_checksum) 412219351Spjd return (true); 413219354Spjd if (res0->hr_compression != res1->hr_compression) 414219354Spjd return (true); 415210886Spjd if (res0->hr_timeout != res1->hr_timeout) 416210886Spjd return (true); 417211886Spjd if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 418211886Spjd return (true); 419210886Spjd return (false); 420210886Spjd} 421210886Spjd 422204076Spjdstatic void 423217784Spjdresource_reload(const struct hast_resource *res) 424217784Spjd{ 425217784Spjd struct nv *nvin, *nvout; 426217784Spjd int error; 427217784Spjd 428218138Spjd PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 429217784Spjd 430217784Spjd nvout = nv_alloc(); 431221076Strociny nv_add_uint8(nvout, CONTROL_RELOAD, "cmd"); 432217784Spjd nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr"); 433219818Spjd nv_add_string(nvout, res->hr_sourceaddr, "sourceaddr"); 434217784Spjd nv_add_int32(nvout, (int32_t)res->hr_replication, "replication"); 435219351Spjd nv_add_int32(nvout, (int32_t)res->hr_checksum, "checksum"); 436219354Spjd nv_add_int32(nvout, (int32_t)res->hr_compression, "compression"); 437217784Spjd nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout"); 438217784Spjd nv_add_string(nvout, res->hr_exec, "exec"); 439217784Spjd if (nv_error(nvout) != 0) { 440217784Spjd nv_free(nvout); 441217784Spjd pjdlog_error("Unable to allocate header for reload message."); 442217784Spjd return; 443217784Spjd } 444217784Spjd if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) { 445217784Spjd pjdlog_errno(LOG_ERR, "Unable to send reload message"); 446217784Spjd nv_free(nvout); 447217784Spjd return; 448217784Spjd } 449217784Spjd nv_free(nvout); 450217784Spjd 451217784Spjd /* Receive response. */ 452217784Spjd if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) { 453217784Spjd pjdlog_errno(LOG_ERR, "Unable to receive reload reply"); 454217784Spjd return; 455217784Spjd } 456217784Spjd error = nv_get_int16(nvin, "error"); 457217784Spjd nv_free(nvin); 458217784Spjd if (error != 0) { 459217784Spjd pjdlog_common(LOG_ERR, 0, error, "Reload failed"); 460217784Spjd return; 461217784Spjd } 462217784Spjd} 463217784Spjd 464217784Spjdstatic void 465204076Spjdhastd_reload(void) 466204076Spjd{ 467210886Spjd struct hastd_config *newcfg; 468210886Spjd struct hast_resource *nres, *cres, *tres; 469222108Spjd struct hastd_listen *nlst, *clst; 470222108Spjd unsigned int nlisten; 471210886Spjd uint8_t role; 472204076Spjd 473210886Spjd pjdlog_info("Reloading configuration..."); 474210886Spjd 475210886Spjd newcfg = yy_config_parse(cfgpath, false); 476210886Spjd if (newcfg == NULL) 477210886Spjd goto failed; 478210886Spjd 479210886Spjd /* 480210886Spjd * Check if control address has changed. 481210886Spjd */ 482210886Spjd if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 483210886Spjd if (proto_server(newcfg->hc_controladdr, 484210886Spjd &newcfg->hc_controlconn) < 0) { 485210886Spjd pjdlog_errno(LOG_ERR, 486210886Spjd "Unable to listen on control address %s", 487210886Spjd newcfg->hc_controladdr); 488210886Spjd goto failed; 489210886Spjd } 490210886Spjd } 491210886Spjd /* 492222108Spjd * Check if any listen address has changed. 493210886Spjd */ 494222108Spjd nlisten = 0; 495222108Spjd TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) { 496222108Spjd TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) { 497222108Spjd if (strcmp(nlst->hl_addr, clst->hl_addr) == 0) 498222108Spjd break; 499210886Spjd } 500222108Spjd if (clst != NULL && clst->hl_conn != NULL) { 501222108Spjd pjdlog_info("Keep listening on address %s.", 502222108Spjd nlst->hl_addr); 503222108Spjd nlst->hl_conn = clst->hl_conn; 504222108Spjd nlisten++; 505222108Spjd } else if (proto_server(nlst->hl_addr, &nlst->hl_conn) == 0) { 506222108Spjd pjdlog_info("Listening on new address %s.", 507222108Spjd nlst->hl_addr); 508222108Spjd nlisten++; 509222108Spjd } else { 510222108Spjd pjdlog_errno(LOG_WARNING, 511222108Spjd "Unable to listen on address %s", nlst->hl_addr); 512222108Spjd } 513210886Spjd } 514222108Spjd if (nlisten == 0) { 515222108Spjd pjdlog_error("No addresses to listen on."); 516222108Spjd goto failed; 517222108Spjd } 518222108Spjd 519222108Spjd /* No failures from now on. */ 520222108Spjd 521210886Spjd /* 522222108Spjd * Switch to new control socket. 523210886Spjd */ 524210886Spjd if (newcfg->hc_controlconn != NULL) { 525210886Spjd pjdlog_info("Control socket changed from %s to %s.", 526210886Spjd cfg->hc_controladdr, newcfg->hc_controladdr); 527210886Spjd proto_close(cfg->hc_controlconn); 528210886Spjd cfg->hc_controlconn = newcfg->hc_controlconn; 529210886Spjd newcfg->hc_controlconn = NULL; 530210886Spjd strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 531210886Spjd sizeof(cfg->hc_controladdr)); 532210886Spjd } 533222108Spjd /* 534222108Spjd * Switch to new listen addresses. Close all that were removed. 535222108Spjd */ 536222108Spjd while ((clst = TAILQ_FIRST(&cfg->hc_listen)) != NULL) { 537222108Spjd TAILQ_FOREACH(nlst, &newcfg->hc_listen, hl_next) { 538222108Spjd if (strcmp(nlst->hl_addr, clst->hl_addr) == 0) 539222108Spjd break; 540222108Spjd } 541222108Spjd if (nlst == NULL && clst->hl_conn != NULL) { 542222108Spjd proto_close(clst->hl_conn); 543222108Spjd pjdlog_info("No longer listening on address %s.", 544222108Spjd clst->hl_addr); 545222108Spjd } 546222108Spjd TAILQ_REMOVE(&cfg->hc_listen, clst, hl_next); 547222108Spjd free(clst); 548210886Spjd } 549222108Spjd TAILQ_CONCAT(&cfg->hc_listen, &newcfg->hc_listen, hl_next); 550210886Spjd 551210886Spjd /* 552210886Spjd * Stop and remove resources that were removed from the configuration. 553210886Spjd */ 554210886Spjd TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 555210886Spjd TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 556210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 557210886Spjd break; 558210886Spjd } 559210886Spjd if (nres == NULL) { 560210886Spjd control_set_role(cres, HAST_ROLE_INIT); 561210886Spjd TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 562210886Spjd pjdlog_info("Resource %s removed.", cres->hr_name); 563210886Spjd free(cres); 564210886Spjd } 565210886Spjd } 566210886Spjd /* 567210886Spjd * Move new resources to the current configuration. 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 } 574210886Spjd if (cres == NULL) { 575210886Spjd TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 576210886Spjd TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 577210886Spjd pjdlog_info("Resource %s added.", nres->hr_name); 578210886Spjd } 579210886Spjd } 580210886Spjd /* 581210886Spjd * Deal with modified resources. 582210886Spjd * Depending on what has changed exactly we might want to perform 583210886Spjd * different actions. 584210886Spjd * 585210886Spjd * We do full resource restart in the following situations: 586210886Spjd * Resource role is INIT or SECONDARY. 587210886Spjd * Resource role is PRIMARY and path to local component or provider 588210886Spjd * name has changed. 589210886Spjd * In case of PRIMARY, the worker process will be killed and restarted, 590210886Spjd * which also means removing /dev/hast/<name> provider and 591210886Spjd * recreating it. 592210886Spjd * 593210886Spjd * We do just reload (send SIGHUP to worker process) if we act as 594217729Spjd * PRIMARY, but only if remote address, replication mode, timeout or 595217729Spjd * execution path has changed. For those, there is no need to restart 596217729Spjd * worker process. 597210886Spjd * If PRIMARY receives SIGHUP, it will reconnect if remote address or 598210886Spjd * replication mode has changed or simply set new timeout if only 599210886Spjd * timeout has changed. 600210886Spjd */ 601210886Spjd TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 602210886Spjd TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 603210886Spjd if (strcmp(cres->hr_name, nres->hr_name) == 0) 604210886Spjd break; 605210886Spjd } 606218138Spjd PJDLOG_ASSERT(cres != NULL); 607210886Spjd if (resource_needs_restart(cres, nres)) { 608210886Spjd pjdlog_info("Resource %s configuration was modified, restarting it.", 609210886Spjd cres->hr_name); 610210886Spjd role = cres->hr_role; 611210886Spjd control_set_role(cres, HAST_ROLE_INIT); 612210886Spjd TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 613210886Spjd free(cres); 614210886Spjd TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 615210886Spjd TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 616210886Spjd control_set_role(nres, role); 617210886Spjd } else if (resource_needs_reload(cres, nres)) { 618210886Spjd pjdlog_info("Resource %s configuration was modified, reloading it.", 619210886Spjd cres->hr_name); 620210886Spjd strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 621210886Spjd sizeof(cres->hr_remoteaddr)); 622219818Spjd strlcpy(cres->hr_sourceaddr, nres->hr_sourceaddr, 623219818Spjd sizeof(cres->hr_sourceaddr)); 624210886Spjd cres->hr_replication = nres->hr_replication; 625219351Spjd cres->hr_checksum = nres->hr_checksum; 626219354Spjd cres->hr_compression = nres->hr_compression; 627210886Spjd cres->hr_timeout = nres->hr_timeout; 628217729Spjd strlcpy(cres->hr_exec, nres->hr_exec, 629217729Spjd sizeof(cres->hr_exec)); 630217784Spjd if (cres->hr_workerpid != 0) 631217784Spjd resource_reload(cres); 632210886Spjd } 633210886Spjd } 634210886Spjd 635210886Spjd yy_config_free(newcfg); 636210886Spjd pjdlog_info("Configuration reloaded successfully."); 637210886Spjd return; 638210886Spjdfailed: 639210886Spjd if (newcfg != NULL) { 640210886Spjd if (newcfg->hc_controlconn != NULL) 641210886Spjd proto_close(newcfg->hc_controlconn); 642222108Spjd while ((nlst = TAILQ_FIRST(&newcfg->hc_listen)) != NULL) { 643222108Spjd if (nlst->hl_conn != NULL) { 644222108Spjd TAILQ_FOREACH(clst, &cfg->hc_listen, hl_next) { 645222108Spjd if (strcmp(nlst->hl_addr, 646222108Spjd clst->hl_addr) == 0) { 647222108Spjd break; 648222108Spjd } 649222108Spjd } 650222108Spjd if (clst == NULL || clst->hl_conn == NULL) 651222108Spjd proto_close(nlst->hl_conn); 652222108Spjd } 653222108Spjd TAILQ_REMOVE(&newcfg->hc_listen, nlst, hl_next); 654222108Spjd free(nlst); 655222108Spjd } 656210886Spjd yy_config_free(newcfg); 657210886Spjd } 658210886Spjd pjdlog_warning("Configuration not reloaded."); 659204076Spjd} 660204076Spjd 661204076Spjdstatic void 662211899Spjdterminate_workers(void) 663211899Spjd{ 664211899Spjd struct hast_resource *res; 665211899Spjd 666211899Spjd pjdlog_info("Termination signal received, exiting."); 667211899Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 668211899Spjd if (res->hr_workerpid == 0) 669211899Spjd continue; 670211899Spjd pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 671211899Spjd res->hr_name, role2str(res->hr_role), res->hr_workerpid); 672211899Spjd if (kill(res->hr_workerpid, SIGTERM) == 0) 673211899Spjd continue; 674211899Spjd pjdlog_errno(LOG_WARNING, 675211899Spjd "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 676211899Spjd res->hr_name, role2str(res->hr_role), res->hr_workerpid); 677211899Spjd } 678211899Spjd} 679211899Spjd 680211899Spjdstatic void 681222108Spjdlisten_accept(struct hastd_listen *lst) 682204076Spjd{ 683204076Spjd struct hast_resource *res; 684204076Spjd struct proto_conn *conn; 685204076Spjd struct nv *nvin, *nvout, *nverr; 686204076Spjd const char *resname; 687204076Spjd const unsigned char *token; 688204076Spjd char laddr[256], raddr[256]; 689204076Spjd size_t size; 690204076Spjd pid_t pid; 691204076Spjd int status; 692204076Spjd 693222108Spjd proto_local_address(lst->hl_conn, laddr, sizeof(laddr)); 694204076Spjd pjdlog_debug(1, "Accepting connection to %s.", laddr); 695204076Spjd 696222108Spjd if (proto_accept(lst->hl_conn, &conn) < 0) { 697204076Spjd pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 698204076Spjd return; 699204076Spjd } 700204076Spjd 701204076Spjd proto_local_address(conn, laddr, sizeof(laddr)); 702204076Spjd proto_remote_address(conn, raddr, sizeof(raddr)); 703209185Spjd pjdlog_info("Connection from %s to %s.", raddr, laddr); 704204076Spjd 705207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 706207371Spjd if (proto_timeout(conn, HAST_TIMEOUT) < 0) 707207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 708207371Spjd 709204076Spjd nvin = nvout = nverr = NULL; 710204076Spjd 711204076Spjd /* 712204076Spjd * Before receiving any data see if remote host have access to any 713204076Spjd * resource. 714204076Spjd */ 715204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 716204076Spjd if (proto_address_match(conn, res->hr_remoteaddr)) 717204076Spjd break; 718204076Spjd } 719204076Spjd if (res == NULL) { 720204076Spjd pjdlog_error("Client %s isn't known.", raddr); 721204076Spjd goto close; 722204076Spjd } 723204076Spjd /* Ok, remote host can access at least one resource. */ 724204076Spjd 725204076Spjd if (hast_proto_recv_hdr(conn, &nvin) < 0) { 726204076Spjd pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 727204076Spjd raddr); 728204076Spjd goto close; 729204076Spjd } 730204076Spjd 731204076Spjd resname = nv_get_string(nvin, "resource"); 732204076Spjd if (resname == NULL) { 733204076Spjd pjdlog_error("No 'resource' field in the header received from %s.", 734204076Spjd raddr); 735204076Spjd goto close; 736204076Spjd } 737204076Spjd pjdlog_debug(2, "%s: resource=%s", raddr, resname); 738204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 739204076Spjd /* 740204076Spjd * NULL token means that this is first conection. 741204076Spjd */ 742204076Spjd if (token != NULL && size != sizeof(res->hr_token)) { 743204076Spjd pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 744204076Spjd raddr, sizeof(res->hr_token), size); 745204076Spjd goto close; 746204076Spjd } 747204076Spjd 748204076Spjd /* 749204076Spjd * From now on we want to send errors to the remote node. 750204076Spjd */ 751204076Spjd nverr = nv_alloc(); 752204076Spjd 753204076Spjd /* Find resource related to this connection. */ 754204076Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 755204076Spjd if (strcmp(resname, res->hr_name) == 0) 756204076Spjd break; 757204076Spjd } 758204076Spjd /* Have we found the resource? */ 759204076Spjd if (res == NULL) { 760204076Spjd pjdlog_error("No resource '%s' as requested by %s.", 761204076Spjd resname, raddr); 762204076Spjd nv_add_stringf(nverr, "errmsg", "Resource not configured."); 763204076Spjd goto fail; 764204076Spjd } 765204076Spjd 766204076Spjd /* Now that we know resource name setup log prefix. */ 767204076Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 768204076Spjd 769204076Spjd /* Does the remote host have access to this resource? */ 770204076Spjd if (!proto_address_match(conn, res->hr_remoteaddr)) { 771204076Spjd pjdlog_error("Client %s has no access to the resource.", raddr); 772204076Spjd nv_add_stringf(nverr, "errmsg", "No access to the resource."); 773204076Spjd goto fail; 774204076Spjd } 775204076Spjd /* Is the resource marked as secondary? */ 776204076Spjd if (res->hr_role != HAST_ROLE_SECONDARY) { 777220890Spjd pjdlog_warning("We act as %s for the resource and not as %s as requested by %s.", 778204076Spjd role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 779204076Spjd raddr); 780204076Spjd nv_add_stringf(nverr, "errmsg", 781204076Spjd "Remote node acts as %s for the resource and not as %s.", 782204076Spjd role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 783220898Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 784220898Spjd /* 785220898Spjd * If we act as primary request the other side to wait 786220899Spjd * for us a bit, as we might be finishing cleanups. 787220898Spjd */ 788220898Spjd nv_add_uint8(nverr, 1, "wait"); 789220898Spjd } 790204076Spjd goto fail; 791204076Spjd } 792204076Spjd /* Does token (if exists) match? */ 793204076Spjd if (token != NULL && memcmp(token, res->hr_token, 794204076Spjd sizeof(res->hr_token)) != 0) { 795204076Spjd pjdlog_error("Token received from %s doesn't match.", raddr); 796209185Spjd nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 797204076Spjd goto fail; 798204076Spjd } 799204076Spjd /* 800204076Spjd * If there is no token, but we have half-open connection 801204076Spjd * (only remotein) or full connection (worker process is running) 802204076Spjd * we have to cancel those and accept the new connection. 803204076Spjd */ 804204076Spjd if (token == NULL) { 805218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 806204076Spjd pjdlog_debug(1, "Initial connection from %s.", raddr); 807204076Spjd if (res->hr_workerpid != 0) { 808218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 809204076Spjd pjdlog_debug(1, 810204076Spjd "Worker process exists (pid=%u), stopping it.", 811204076Spjd (unsigned int)res->hr_workerpid); 812204076Spjd /* Stop child process. */ 813204076Spjd if (kill(res->hr_workerpid, SIGINT) < 0) { 814204076Spjd pjdlog_errno(LOG_ERR, 815204076Spjd "Unable to stop worker process (pid=%u)", 816204076Spjd (unsigned int)res->hr_workerpid); 817204076Spjd /* 818204076Spjd * Other than logging the problem we 819204076Spjd * ignore it - nothing smart to do. 820204076Spjd */ 821204076Spjd } 822204076Spjd /* Wait for it to exit. */ 823204076Spjd else if ((pid = waitpid(res->hr_workerpid, 824204076Spjd &status, 0)) != res->hr_workerpid) { 825207372Spjd /* We can only log the problem. */ 826204076Spjd pjdlog_errno(LOG_ERR, 827204076Spjd "Waiting for worker process (pid=%u) failed", 828204076Spjd (unsigned int)res->hr_workerpid); 829204076Spjd } else { 830207372Spjd child_exit_log(res->hr_workerpid, status); 831204076Spjd } 832213006Spjd child_cleanup(res); 833204076Spjd } else if (res->hr_remotein != NULL) { 834204076Spjd char oaddr[256]; 835204076Spjd 836213981Spjd proto_remote_address(res->hr_remotein, oaddr, 837213981Spjd sizeof(oaddr)); 838204076Spjd pjdlog_debug(1, 839204076Spjd "Canceling half-open connection from %s on connection from %s.", 840204076Spjd oaddr, raddr); 841204076Spjd proto_close(res->hr_remotein); 842204076Spjd res->hr_remotein = NULL; 843204076Spjd } 844204076Spjd } 845204076Spjd 846204076Spjd /* 847204076Spjd * Checks and cleanups are done. 848204076Spjd */ 849204076Spjd 850204076Spjd if (token == NULL) { 851204076Spjd arc4random_buf(res->hr_token, sizeof(res->hr_token)); 852204076Spjd nvout = nv_alloc(); 853204076Spjd nv_add_uint8_array(nvout, res->hr_token, 854204076Spjd sizeof(res->hr_token), "token"); 855204076Spjd if (nv_error(nvout) != 0) { 856204076Spjd pjdlog_common(LOG_ERR, 0, nv_error(nvout), 857204076Spjd "Unable to prepare return header for %s", raddr); 858204076Spjd nv_add_stringf(nverr, "errmsg", 859204076Spjd "Remote node was unable to prepare return header: %s.", 860204076Spjd strerror(nv_error(nvout))); 861204076Spjd goto fail; 862204076Spjd } 863204076Spjd if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) { 864204076Spjd int error = errno; 865204076Spjd 866204076Spjd pjdlog_errno(LOG_ERR, "Unable to send response to %s", 867204076Spjd raddr); 868204076Spjd nv_add_stringf(nverr, "errmsg", 869204076Spjd "Remote node was unable to send response: %s.", 870204076Spjd strerror(error)); 871204076Spjd goto fail; 872204076Spjd } 873204076Spjd res->hr_remotein = conn; 874204076Spjd pjdlog_debug(1, "Incoming connection from %s configured.", 875204076Spjd raddr); 876204076Spjd } else { 877204076Spjd res->hr_remoteout = conn; 878204076Spjd pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 879204076Spjd hastd_secondary(res, nvin); 880204076Spjd } 881204076Spjd nv_free(nvin); 882204076Spjd nv_free(nvout); 883204076Spjd nv_free(nverr); 884204076Spjd pjdlog_prefix_set("%s", ""); 885204076Spjd return; 886204076Spjdfail: 887204076Spjd if (nv_error(nverr) != 0) { 888204076Spjd pjdlog_common(LOG_ERR, 0, nv_error(nverr), 889204076Spjd "Unable to prepare error header for %s", raddr); 890204076Spjd goto close; 891204076Spjd } 892204076Spjd if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) { 893204076Spjd pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 894204076Spjd goto close; 895204076Spjd } 896204076Spjdclose: 897204076Spjd if (nvin != NULL) 898204076Spjd nv_free(nvin); 899204076Spjd if (nvout != NULL) 900204076Spjd nv_free(nvout); 901204076Spjd if (nverr != NULL) 902204076Spjd nv_free(nverr); 903204076Spjd proto_close(conn); 904204076Spjd pjdlog_prefix_set("%s", ""); 905204076Spjd} 906204076Spjd 907204076Spjdstatic void 908218218Spjdconnection_migrate(struct hast_resource *res) 909218218Spjd{ 910218218Spjd struct proto_conn *conn; 911218218Spjd int16_t val = 0; 912218218Spjd 913219814Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 914219814Spjd 915219900Spjd PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 916219900Spjd 917218218Spjd if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) { 918218218Spjd pjdlog_errno(LOG_WARNING, 919218218Spjd "Unable to receive connection command"); 920218218Spjd return; 921218218Spjd } 922219818Spjd if (proto_client(res->hr_sourceaddr[0] != '\0' ? res->hr_sourceaddr : NULL, 923219818Spjd res->hr_remoteaddr, &conn) < 0) { 924218218Spjd val = errno; 925218218Spjd pjdlog_errno(LOG_WARNING, 926218218Spjd "Unable to create outgoing connection to %s", 927218218Spjd res->hr_remoteaddr); 928218218Spjd goto out; 929218218Spjd } 930218218Spjd if (proto_connect(conn, -1) < 0) { 931218218Spjd val = errno; 932218218Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 933218218Spjd res->hr_remoteaddr); 934218218Spjd proto_close(conn); 935218218Spjd goto out; 936218218Spjd } 937218218Spjd val = 0; 938218218Spjdout: 939218218Spjd if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) { 940218218Spjd pjdlog_errno(LOG_WARNING, 941218218Spjd "Unable to send reply to connection request"); 942218218Spjd } 943218218Spjd if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0) 944218218Spjd pjdlog_errno(LOG_WARNING, "Unable to send connection"); 945219814Spjd 946219814Spjd pjdlog_prefix_set("%s", ""); 947218218Spjd} 948218218Spjd 949218218Spjdstatic void 950219837Spjdcheck_signals(void) 951204076Spjd{ 952213009Spjd struct timespec sigtimeout; 953213009Spjd sigset_t mask; 954219837Spjd int signo; 955204076Spjd 956213009Spjd sigtimeout.tv_sec = 0; 957213009Spjd sigtimeout.tv_nsec = 0; 958211977Spjd 959213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 960213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 961213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 962213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 963213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 964213009Spjd 965219837Spjd while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 966219837Spjd switch (signo) { 967219837Spjd case SIGINT: 968219837Spjd case SIGTERM: 969219837Spjd sigexit_received = true; 970219837Spjd terminate_workers(); 971219837Spjd proto_close(cfg->hc_controlconn); 972219837Spjd exit(EX_OK); 973219837Spjd break; 974219837Spjd case SIGCHLD: 975219837Spjd child_exit(); 976219837Spjd break; 977219837Spjd case SIGHUP: 978219837Spjd hastd_reload(); 979219837Spjd break; 980219837Spjd default: 981219837Spjd PJDLOG_ABORT("Unexpected signal (%d).", signo); 982219837Spjd } 983219837Spjd } 984219837Spjd} 985219837Spjd 986219837Spjdstatic void 987219837Spjdmain_loop(void) 988219837Spjd{ 989219837Spjd struct hast_resource *res; 990222108Spjd struct hastd_listen *lst; 991219837Spjd struct timeval seltimeout; 992219837Spjd int fd, maxfd, ret; 993219837Spjd time_t lastcheck, now; 994219837Spjd fd_set rfds; 995219837Spjd 996219864Spjd lastcheck = time(NULL); 997219837Spjd seltimeout.tv_sec = REPORT_INTERVAL; 998219837Spjd seltimeout.tv_usec = 0; 999219837Spjd 1000204076Spjd for (;;) { 1001219837Spjd check_signals(); 1002204076Spjd 1003209177Spjd /* Setup descriptors for select(2). */ 1004204076Spjd FD_ZERO(&rfds); 1005212038Spjd maxfd = fd = proto_descriptor(cfg->hc_controlconn); 1006218138Spjd PJDLOG_ASSERT(fd >= 0); 1007212038Spjd FD_SET(fd, &rfds); 1008222108Spjd TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) { 1009222108Spjd if (lst->hl_conn == NULL) 1010222108Spjd continue; 1011222108Spjd fd = proto_descriptor(lst->hl_conn); 1012222108Spjd PJDLOG_ASSERT(fd >= 0); 1013222108Spjd FD_SET(fd, &rfds); 1014222108Spjd maxfd = fd > maxfd ? fd : maxfd; 1015222108Spjd } 1016212038Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 1017212038Spjd if (res->hr_event == NULL) 1018212038Spjd continue; 1019212038Spjd fd = proto_descriptor(res->hr_event); 1020218138Spjd PJDLOG_ASSERT(fd >= 0); 1021212038Spjd FD_SET(fd, &rfds); 1022212038Spjd maxfd = fd > maxfd ? fd : maxfd; 1023218218Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 1024218218Spjd /* Only primary workers asks for connections. */ 1025219900Spjd PJDLOG_ASSERT(res->hr_conn != NULL); 1026218218Spjd fd = proto_descriptor(res->hr_conn); 1027218218Spjd PJDLOG_ASSERT(fd >= 0); 1028218218Spjd FD_SET(fd, &rfds); 1029218218Spjd maxfd = fd > maxfd ? fd : maxfd; 1030219900Spjd } else { 1031219900Spjd PJDLOG_ASSERT(res->hr_conn == NULL); 1032218218Spjd } 1033212038Spjd } 1034204076Spjd 1035218138Spjd PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE); 1036213009Spjd ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 1037219813Spjd now = time(NULL); 1038219813Spjd if (lastcheck + REPORT_INTERVAL <= now) { 1039213429Spjd hook_check(); 1040219813Spjd lastcheck = now; 1041219813Spjd } 1042219813Spjd if (ret == 0) { 1043219813Spjd /* 1044219813Spjd * select(2) timed out, so there should be no 1045219813Spjd * descriptors to check. 1046219813Spjd */ 1047219813Spjd continue; 1048219813Spjd } else if (ret == -1) { 1049204076Spjd if (errno == EINTR) 1050204076Spjd continue; 1051204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1052204076Spjd pjdlog_exit(EX_OSERR, "select() failed"); 1053204076Spjd } 1054204076Spjd 1055219837Spjd /* 1056219837Spjd * Check for signals before we do anything to update our 1057219837Spjd * info about terminated workers in the meantime. 1058219837Spjd */ 1059219837Spjd check_signals(); 1060219837Spjd 1061212038Spjd if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 1062204076Spjd control_handle(cfg); 1063222108Spjd TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) { 1064222108Spjd if (lst->hl_conn == NULL) 1065222108Spjd continue; 1066222108Spjd if (FD_ISSET(proto_descriptor(lst->hl_conn), &rfds)) 1067222108Spjd listen_accept(lst); 1068222108Spjd } 1069212038Spjd TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 1070212038Spjd if (res->hr_event == NULL) 1071212038Spjd continue; 1072212038Spjd if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 1073212038Spjd if (event_recv(res) == 0) 1074212038Spjd continue; 1075212038Spjd /* The worker process exited? */ 1076212038Spjd proto_close(res->hr_event); 1077212038Spjd res->hr_event = NULL; 1078219900Spjd if (res->hr_conn != NULL) { 1079219900Spjd proto_close(res->hr_conn); 1080219900Spjd res->hr_conn = NULL; 1081219900Spjd } 1082218218Spjd continue; 1083212038Spjd } 1084219900Spjd if (res->hr_role == HAST_ROLE_PRIMARY) { 1085219900Spjd PJDLOG_ASSERT(res->hr_conn != NULL); 1086219900Spjd if (FD_ISSET(proto_descriptor(res->hr_conn), 1087219900Spjd &rfds)) { 1088219900Spjd connection_migrate(res); 1089219900Spjd } 1090219900Spjd } else { 1091219900Spjd PJDLOG_ASSERT(res->hr_conn == NULL); 1092218218Spjd } 1093212038Spjd } 1094204076Spjd } 1095204076Spjd} 1096204076Spjd 1097213428Spjdstatic void 1098213428Spjddummy_sighandler(int sig __unused) 1099213428Spjd{ 1100213428Spjd /* Nothing to do. */ 1101213428Spjd} 1102213428Spjd 1103204076Spjdint 1104204076Spjdmain(int argc, char *argv[]) 1105204076Spjd{ 1106222108Spjd struct hastd_listen *lst; 1107204076Spjd const char *pidfile; 1108204076Spjd pid_t otherpid; 1109204076Spjd bool foreground; 1110204076Spjd int debuglevel; 1111213009Spjd sigset_t mask; 1112204076Spjd 1113204076Spjd foreground = false; 1114204076Spjd debuglevel = 0; 1115204076Spjd pidfile = HASTD_PIDFILE; 1116204076Spjd 1117204076Spjd for (;;) { 1118204076Spjd int ch; 1119204076Spjd 1120204076Spjd ch = getopt(argc, argv, "c:dFhP:"); 1121204076Spjd if (ch == -1) 1122204076Spjd break; 1123204076Spjd switch (ch) { 1124204076Spjd case 'c': 1125204076Spjd cfgpath = optarg; 1126204076Spjd break; 1127204076Spjd case 'd': 1128204076Spjd debuglevel++; 1129204076Spjd break; 1130204076Spjd case 'F': 1131204076Spjd foreground = true; 1132204076Spjd break; 1133204076Spjd case 'P': 1134204076Spjd pidfile = optarg; 1135204076Spjd break; 1136204076Spjd case 'h': 1137204076Spjd default: 1138204076Spjd usage(); 1139204076Spjd } 1140204076Spjd } 1141204076Spjd argc -= optind; 1142204076Spjd argv += optind; 1143204076Spjd 1144217965Spjd pjdlog_init(PJDLOG_MODE_STD); 1145204076Spjd pjdlog_debug_set(debuglevel); 1146204076Spjd 1147214273Spjd g_gate_load(); 1148214273Spjd 1149204076Spjd pfh = pidfile_open(pidfile, 0600, &otherpid); 1150204076Spjd if (pfh == NULL) { 1151204076Spjd if (errno == EEXIST) { 1152204076Spjd pjdlog_exitx(EX_TEMPFAIL, 1153204076Spjd "Another hastd is already running, pid: %jd.", 1154204076Spjd (intmax_t)otherpid); 1155204076Spjd } 1156204076Spjd /* If we cannot create pidfile from other reasons, only warn. */ 1157210879Spjd pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile"); 1158204076Spjd } 1159204076Spjd 1160210883Spjd cfg = yy_config_parse(cfgpath, true); 1161218138Spjd PJDLOG_ASSERT(cfg != NULL); 1162204076Spjd 1163213428Spjd /* 1164217307Spjd * Restore default actions for interesting signals in case parent 1165217307Spjd * process (like init(8)) decided to ignore some of them (like SIGHUP). 1166217307Spjd */ 1167217307Spjd PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 1168217307Spjd PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 1169217307Spjd PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 1170217307Spjd /* 1171213428Spjd * Because SIGCHLD is ignored by default, setup dummy handler for it, 1172213428Spjd * so we can mask it. 1173213428Spjd */ 1174213428Spjd PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 1175217307Spjd 1176213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1177213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1178213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1179213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1180213009Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1181213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 1182204076Spjd 1183204076Spjd /* Listen on control address. */ 1184204076Spjd if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) { 1185204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1186204076Spjd pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 1187204076Spjd cfg->hc_controladdr); 1188204076Spjd } 1189204076Spjd /* Listen for remote connections. */ 1190222108Spjd TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) { 1191222108Spjd if (proto_server(lst->hl_addr, &lst->hl_conn) < 0) { 1192222108Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1193222108Spjd pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 1194222108Spjd lst->hl_addr); 1195222108Spjd } 1196204076Spjd } 1197204076Spjd 1198204076Spjd if (!foreground) { 1199204076Spjd if (daemon(0, 0) < 0) { 1200204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 1201204076Spjd pjdlog_exit(EX_OSERR, "Unable to daemonize"); 1202204076Spjd } 1203204076Spjd 1204204076Spjd /* Start logging to syslog. */ 1205204076Spjd pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 1206204076Spjd 1207204076Spjd /* Write PID to a file. */ 1208204076Spjd if (pidfile_write(pfh) < 0) { 1209204076Spjd pjdlog_errno(LOG_WARNING, 1210204076Spjd "Unable to write PID to a file"); 1211204076Spjd } 1212204076Spjd } 1213204076Spjd 1214222108Spjd pjdlog_info("Started successfully, running protocol version %d.", 1215222108Spjd HAST_PROTO_VERSION); 1216222108Spjd 1217222108Spjd pjdlog_debug(1, "Listening on control address %s.", 1218222108Spjd cfg->hc_controladdr); 1219222108Spjd TAILQ_FOREACH(lst, &cfg->hc_listen, hl_next) 1220222108Spjd pjdlog_info("Listening on address %s.", lst->hl_addr); 1221222108Spjd 1222211977Spjd hook_init(); 1223211977Spjd 1224204076Spjd main_loop(); 1225204076Spjd 1226204076Spjd exit(0); 1227204076Spjd} 1228