hastd.c revision 220899
166458Sdfr/*- 266458Sdfr * Copyright (c) 2009-2010 The FreeBSD Foundation 366458Sdfr * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net> 466458Sdfr * All rights reserved. 566458Sdfr * 666458Sdfr * This software was developed by Pawel Jakub Dawidek under sponsorship from 766458Sdfr * the FreeBSD Foundation. 866458Sdfr * 966458Sdfr * Redistribution and use in source and binary forms, with or without 1066458Sdfr * modification, are permitted provided that the following conditions 1166458Sdfr * are met: 1266458Sdfr * 1. Redistributions of source code must retain the above copyright 1366458Sdfr * notice, this list of conditions and the following disclaimer. 1466458Sdfr * 2. Redistributions in binary form must reproduce the above copyright 1566458Sdfr * notice, this list of conditions and the following disclaimer in the 1666458Sdfr * documentation and/or other materials provided with the distribution. 1766458Sdfr * 1866458Sdfr * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 1966458Sdfr * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2066458Sdfr * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2166458Sdfr * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 2266458Sdfr * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2366458Sdfr * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2466458Sdfr * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2566458Sdfr * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2666458Sdfr * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2766458Sdfr * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2866458Sdfr * SUCH DAMAGE. 2966458Sdfr */ 3066458Sdfr 3166458Sdfr#include <sys/cdefs.h> 3266458Sdfr__FBSDID("$FreeBSD: head/sbin/hastd/hastd.c 220899 2011-04-20 18:49:12Z pjd $"); 3366458Sdfr 3466458Sdfr#include <sys/param.h> 3566458Sdfr#include <sys/linker.h> 3666458Sdfr#include <sys/module.h> 3766458Sdfr#include <sys/stat.h> 3866458Sdfr#include <sys/wait.h> 3966458Sdfr 4066458Sdfr#include <err.h> 4166458Sdfr#include <errno.h> 4266458Sdfr#include <libutil.h> 4366458Sdfr#include <signal.h> 4466458Sdfr#include <stdbool.h> 4566458Sdfr#include <stdio.h> 4666458Sdfr#include <stdlib.h> 4766458Sdfr#include <string.h> 4866458Sdfr#include <sysexits.h> 4966458Sdfr#include <time.h> 50117126Sscottl#include <unistd.h> 51117126Sscottl 5266458Sdfr#include <activemap.h> 5366458Sdfr#include <pjdlog.h> 5466458Sdfr 5566458Sdfr#include "control.h" 5666458Sdfr#include "event.h" 5766458Sdfr#include "hast.h" 58146214Snyan#include "hast_proto.h" 5966458Sdfr#include "hastd.h" 6066458Sdfr#include "hooks.h" 6166458Sdfr#include "subr.h" 6266458Sdfr 6366458Sdfr/* Path to configuration file. */ 6466458Sdfrconst char *cfgpath = HAST_CONFIG; 6566458Sdfr/* Hastd configuration. */ 6666458Sdfrstatic struct hastd_config *cfg; 6766458Sdfr/* Was SIGINT or SIGTERM signal received? */ 6866458Sdfrbool sigexit_received = false; 6966458Sdfr/* PID file handle. */ 7066458Sdfrstruct pidfh *pfh; 7166458Sdfr 7266458Sdfr/* How often check for hooks running for too long. */ 7366458Sdfr#define REPORT_INTERVAL 5 7466458Sdfr 7566458Sdfrstatic void 76135262Sphkusage(void) 77135262Sphk{ 7866458Sdfr 7966458Sdfr errx(EX_USAGE, "[-dFh] [-c config] [-P pidfile]"); 8066458Sdfr} 8166458Sdfr 8266458Sdfrstatic void 8366458Sdfrg_gate_load(void) 8466458Sdfr{ 8566458Sdfr 8666458Sdfr if (modfind("g_gate") == -1) { 8766458Sdfr /* Not present in kernel, try loading it. */ 8866458Sdfr if (kldload("geom_gate") == -1 || modfind("g_gate") == -1) { 8966458Sdfr if (errno != EEXIST) { 9066458Sdfr pjdlog_exit(EX_OSERR, 9166458Sdfr "Unable to load geom_gate module"); 9266458Sdfr } 9366458Sdfr } 9466458Sdfr } 95135262Sphk} 9666458Sdfr 9766458Sdfrvoid 98135262Sphkdescriptors_cleanup(struct hast_resource *res) 9966458Sdfr{ 10066458Sdfr struct hast_resource *tres; 10166458Sdfr 10266458Sdfr TAILQ_FOREACH(tres, &cfg->hc_resources, hr_next) { 10366458Sdfr if (tres == res) { 10466458Sdfr PJDLOG_VERIFY(res->hr_role == HAST_ROLE_SECONDARY || 10566458Sdfr (res->hr_remotein == NULL && 10666458Sdfr res->hr_remoteout == NULL)); 10766458Sdfr continue; 10866458Sdfr } 109177215Simp if (tres->hr_remotein != NULL) 110117126Sscottl proto_close(tres->hr_remotein); 111117126Sscottl if (tres->hr_remoteout != NULL) 11266458Sdfr proto_close(tres->hr_remoteout); 113135262Sphk if (tres->hr_ctrl != NULL) 11466458Sdfr proto_close(tres->hr_ctrl); 11566458Sdfr if (tres->hr_event != NULL) 11666458Sdfr proto_close(tres->hr_event); 117135262Sphk if (tres->hr_conn != NULL) 11866458Sdfr proto_close(tres->hr_conn); 11966458Sdfr } 120135262Sphk if (cfg->hc_controlin != NULL) 12166458Sdfr proto_close(cfg->hc_controlin); 12266458Sdfr proto_close(cfg->hc_controlconn); 12366458Sdfr proto_close(cfg->hc_listenconn); 12466458Sdfr (void)pidfile_close(pfh); 12566458Sdfr hook_fini(); 12666458Sdfr pjdlog_fini(); 12766458Sdfr} 12866458Sdfr 12966458Sdfrstatic const char * 13066458Sdfrdtype2str(mode_t mode) 13166458Sdfr{ 13266458Sdfr 13366458Sdfr if (S_ISBLK(mode)) 13466458Sdfr return ("block device"); 13566458Sdfr else if (S_ISCHR(mode)) 13666458Sdfr return ("character device"); 13766458Sdfr else if (S_ISDIR(mode)) 13866458Sdfr return ("directory"); 13966458Sdfr else if (S_ISFIFO(mode)) 14066458Sdfr return ("pipe or FIFO"); 14166458Sdfr else if (S_ISLNK(mode)) 14266458Sdfr return ("symbolic link"); 14366458Sdfr else if (S_ISREG(mode)) 14466458Sdfr return ("regular file"); 14566458Sdfr else if (S_ISSOCK(mode)) 14666458Sdfr return ("socket"); 14766458Sdfr else if (S_ISWHT(mode)) 14866458Sdfr return ("whiteout"); 14966458Sdfr else 15066458Sdfr return ("unknown"); 15166458Sdfr} 15266458Sdfr 15366458Sdfrvoid 15466458Sdfrdescriptors_assert(const struct hast_resource *res, int pjdlogmode) 15566458Sdfr{ 15666458Sdfr char msg[256]; 15766458Sdfr struct stat sb; 15866458Sdfr long maxfd; 15966458Sdfr bool isopen; 16066458Sdfr mode_t mode; 16166458Sdfr int fd; 16266458Sdfr 16366458Sdfr /* 16466458Sdfr * At this point descriptor to syslog socket is closed, so if we want 16566458Sdfr * to log assertion message, we have to first store it in 'msg' local 16666458Sdfr * buffer and then open syslog socket and log it. 16766458Sdfr */ 16866458Sdfr msg[0] = '\0'; 16966458Sdfr 17066458Sdfr maxfd = sysconf(_SC_OPEN_MAX); 17166458Sdfr if (maxfd < 0) { 17266458Sdfr pjdlog_init(pjdlogmode); 17366458Sdfr pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 17466458Sdfr role2str(res->hr_role)); 17566458Sdfr pjdlog_errno(LOG_WARNING, "sysconf(_SC_OPEN_MAX) failed"); 17666458Sdfr pjdlog_fini(); 17766458Sdfr maxfd = 16384; 17866458Sdfr } 17966458Sdfr for (fd = 0; fd <= maxfd; fd++) { 18066458Sdfr if (fstat(fd, &sb) == 0) { 18166458Sdfr isopen = true; 18266458Sdfr mode = sb.st_mode; 18366458Sdfr } else if (errno == EBADF) { 18466458Sdfr isopen = false; 18566458Sdfr mode = 0; 18666458Sdfr } else { 18766458Sdfr (void)snprintf(msg, sizeof(msg), 18866458Sdfr "Unable to fstat descriptor %d: %s", fd, 18966458Sdfr strerror(errno)); 19066458Sdfr break; 19166458Sdfr } 19266458Sdfr if (fd == STDIN_FILENO || fd == STDOUT_FILENO || 19366458Sdfr fd == STDERR_FILENO) { 19466458Sdfr if (!isopen) { 19566458Sdfr (void)snprintf(msg, sizeof(msg), 19666458Sdfr "Descriptor %d (%s) is closed, but should be open.", 19766458Sdfr fd, (fd == STDIN_FILENO ? "stdin" : 19866458Sdfr (fd == STDOUT_FILENO ? "stdout" : "stderr"))); 19966458Sdfr break; 20066458Sdfr } 20166458Sdfr } else if (fd == proto_descriptor(res->hr_event)) { 20266458Sdfr if (!isopen) { 20366458Sdfr (void)snprintf(msg, sizeof(msg), 20466458Sdfr "Descriptor %d (event) is closed, but should be open.", 20566458Sdfr fd); 20666458Sdfr break; 20766458Sdfr } 20866458Sdfr if (!S_ISSOCK(mode)) { 20966458Sdfr (void)snprintf(msg, sizeof(msg), 21066458Sdfr "Descriptor %d (event) is %s, but should be %s.", 21166458Sdfr fd, dtype2str(mode), dtype2str(S_IFSOCK)); 21292676Speter break; 21366458Sdfr } 21492676Speter } else if (fd == proto_descriptor(res->hr_ctrl)) { 21566458Sdfr if (!isopen) { 21666458Sdfr (void)snprintf(msg, sizeof(msg), 21766458Sdfr "Descriptor %d (ctrl) is closed, but should be open.", 21866458Sdfr fd); 21966458Sdfr break; 22066458Sdfr } 22166458Sdfr if (!S_ISSOCK(mode)) { 22266458Sdfr (void)snprintf(msg, sizeof(msg), 22366458Sdfr "Descriptor %d (ctrl) is %s, but should be %s.", 22466458Sdfr fd, dtype2str(mode), dtype2str(S_IFSOCK)); 22566458Sdfr break; 22695710Speter } 22766458Sdfr } else if (res->hr_role == HAST_ROLE_PRIMARY && 22866458Sdfr fd == proto_descriptor(res->hr_conn)) { 22966458Sdfr if (!isopen) { 23066458Sdfr (void)snprintf(msg, sizeof(msg), 23166458Sdfr "Descriptor %d (conn) is closed, but should be open.", 23266458Sdfr fd); 23366458Sdfr break; 23466458Sdfr } 23566458Sdfr if (!S_ISSOCK(mode)) { 23666458Sdfr (void)snprintf(msg, sizeof(msg), 23766458Sdfr "Descriptor %d (conn) is %s, but should be %s.", 23866458Sdfr fd, dtype2str(mode), dtype2str(S_IFSOCK)); 23966458Sdfr break; 24066458Sdfr } 24166458Sdfr } else if (res->hr_role == HAST_ROLE_SECONDARY && 24266458Sdfr res->hr_conn != NULL && 24366458Sdfr fd == proto_descriptor(res->hr_conn)) { 24466458Sdfr if (isopen) { 24566458Sdfr (void)snprintf(msg, sizeof(msg), 24666458Sdfr "Descriptor %d (conn) is open, but should be closed.", 24766458Sdfr fd); 24866458Sdfr break; 24966458Sdfr } 25066458Sdfr } else if (res->hr_role == HAST_ROLE_SECONDARY && 25166458Sdfr fd == proto_descriptor(res->hr_remotein)) { 25266458Sdfr if (!isopen) { 25366458Sdfr (void)snprintf(msg, sizeof(msg), 25466458Sdfr "Descriptor %d (remote in) is closed, but should be open.", 25566458Sdfr fd); 25666458Sdfr break; 25766458Sdfr } 25866458Sdfr if (!S_ISSOCK(mode)) { 25966458Sdfr (void)snprintf(msg, sizeof(msg), 26066458Sdfr "Descriptor %d (remote in) is %s, but should be %s.", 26166458Sdfr fd, dtype2str(mode), dtype2str(S_IFSOCK)); 26266458Sdfr break; 26366458Sdfr } 26466458Sdfr } else if (res->hr_role == HAST_ROLE_SECONDARY && 26566458Sdfr fd == proto_descriptor(res->hr_remoteout)) { 26666458Sdfr if (!isopen) { 26766458Sdfr (void)snprintf(msg, sizeof(msg), 26866458Sdfr "Descriptor %d (remote out) is closed, but should be open.", 26966458Sdfr fd); 27066458Sdfr break; 27166458Sdfr } 27266458Sdfr if (!S_ISSOCK(mode)) { 27366458Sdfr (void)snprintf(msg, sizeof(msg), 27466458Sdfr "Descriptor %d (remote out) is %s, but should be %s.", 27566458Sdfr fd, dtype2str(mode), dtype2str(S_IFSOCK)); 27666458Sdfr break; 27766458Sdfr } 27866458Sdfr } else { 27966458Sdfr if (isopen) { 28066458Sdfr (void)snprintf(msg, sizeof(msg), 28166458Sdfr "Descriptor %d is open (%s), but should be closed.", 28266458Sdfr fd, dtype2str(mode)); 28366458Sdfr break; 28466458Sdfr } 28566458Sdfr } 28666458Sdfr } 28766458Sdfr if (msg[0] != '\0') { 28866458Sdfr pjdlog_init(pjdlogmode); 28966458Sdfr pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 29066458Sdfr role2str(res->hr_role)); 29166458Sdfr PJDLOG_ABORT("%s", msg); 29266458Sdfr } 29366458Sdfr} 29466458Sdfr 29566458Sdfrstatic void 29666458Sdfrchild_exit_log(unsigned int pid, int status) 29766458Sdfr{ 29866458Sdfr 29966458Sdfr if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 30066458Sdfr pjdlog_debug(1, "Worker process exited gracefully (pid=%u).", 30166458Sdfr pid); 30266458Sdfr } else if (WIFSIGNALED(status)) { 30366458Sdfr pjdlog_error("Worker process killed (pid=%u, signal=%d).", 30466458Sdfr pid, WTERMSIG(status)); 30566458Sdfr } else { 30666458Sdfr pjdlog_error("Worker process exited ungracefully (pid=%u, exitcode=%d).", 30766458Sdfr pid, WIFEXITED(status) ? WEXITSTATUS(status) : -1); 30866458Sdfr } 30966458Sdfr} 31066458Sdfr 31166458Sdfrstatic void 31266458Sdfrchild_exit(void) 31366458Sdfr{ 31466458Sdfr struct hast_resource *res; 31566458Sdfr int status; 31666458Sdfr pid_t pid; 31766458Sdfr 31866458Sdfr while ((pid = wait3(&status, WNOHANG, NULL)) > 0) { 31966458Sdfr /* Find resource related to the process that just exited. */ 32066458Sdfr TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 32166458Sdfr if (pid == res->hr_workerpid) 32266458Sdfr break; 32366458Sdfr } 32466458Sdfr if (res == NULL) { 32566458Sdfr /* 32666458Sdfr * This can happen when new connection arrives and we 32766458Sdfr * cancel child responsible for the old one or if this 32866458Sdfr * was hook which we executed. 32966458Sdfr */ 33066458Sdfr hook_check_one(pid, status); 33166458Sdfr continue; 33266458Sdfr } 33366458Sdfr pjdlog_prefix_set("[%s] (%s) ", res->hr_name, 334171312Smarcel role2str(res->hr_role)); 335135262Sphk child_exit_log(pid, status); 33666458Sdfr child_cleanup(res); 33766458Sdfr if (res->hr_role == HAST_ROLE_PRIMARY) { 33866458Sdfr /* 33966458Sdfr * Restart child process if it was killed by signal 34066458Sdfr * or exited because of temporary problem. 34166458Sdfr */ 34266458Sdfr if (WIFSIGNALED(status) || 34366458Sdfr (WIFEXITED(status) && 34466458Sdfr WEXITSTATUS(status) == EX_TEMPFAIL)) { 34566458Sdfr sleep(1); 34666458Sdfr pjdlog_info("Restarting worker process."); 34766458Sdfr hastd_primary(res); 34866458Sdfr } else { 34966458Sdfr res->hr_role = HAST_ROLE_INIT; 35066458Sdfr pjdlog_info("Changing resource role back to %s.", 35166458Sdfr role2str(res->hr_role)); 35266458Sdfr } 35366458Sdfr } 35466458Sdfr pjdlog_prefix_set("%s", ""); 35566458Sdfr } 35666458Sdfr} 35766458Sdfr 35866458Sdfrstatic bool 35966458Sdfrresource_needs_restart(const struct hast_resource *res0, 36066458Sdfr const struct hast_resource *res1) 36166458Sdfr{ 36266458Sdfr 36366458Sdfr PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 36466458Sdfr 36566458Sdfr if (strcmp(res0->hr_provname, res1->hr_provname) != 0) 36666458Sdfr return (true); 36766458Sdfr if (strcmp(res0->hr_localpath, res1->hr_localpath) != 0) 36866458Sdfr return (true); 36966458Sdfr if (res0->hr_role == HAST_ROLE_INIT || 37066458Sdfr res0->hr_role == HAST_ROLE_SECONDARY) { 37166458Sdfr if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 37266458Sdfr return (true); 37366458Sdfr if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0) 37466458Sdfr return (true); 37566458Sdfr if (res0->hr_replication != res1->hr_replication) 37666458Sdfr return (true); 37766458Sdfr if (res0->hr_checksum != res1->hr_checksum) 37866458Sdfr return (true); 37966458Sdfr if (res0->hr_compression != res1->hr_compression) 38066458Sdfr return (true); 38166458Sdfr if (res0->hr_timeout != res1->hr_timeout) 38266458Sdfr return (true); 38366458Sdfr if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 38466458Sdfr return (true); 38566458Sdfr } 38666458Sdfr return (false); 38766458Sdfr} 38866458Sdfr 38966458Sdfrstatic bool 39066458Sdfrresource_needs_reload(const struct hast_resource *res0, 39166458Sdfr const struct hast_resource *res1) 39266458Sdfr{ 39366458Sdfr 39466458Sdfr PJDLOG_ASSERT(strcmp(res0->hr_name, res1->hr_name) == 0); 39566458Sdfr PJDLOG_ASSERT(strcmp(res0->hr_provname, res1->hr_provname) == 0); 39666458Sdfr PJDLOG_ASSERT(strcmp(res0->hr_localpath, res1->hr_localpath) == 0); 39766458Sdfr 39866458Sdfr if (res0->hr_role != HAST_ROLE_PRIMARY) 39966458Sdfr return (false); 40066458Sdfr 40166458Sdfr if (strcmp(res0->hr_remoteaddr, res1->hr_remoteaddr) != 0) 40266458Sdfr return (true); 40366458Sdfr if (strcmp(res0->hr_sourceaddr, res1->hr_sourceaddr) != 0) 40466458Sdfr return (true); 40566458Sdfr if (res0->hr_replication != res1->hr_replication) 40666458Sdfr return (true); 40766458Sdfr if (res0->hr_checksum != res1->hr_checksum) 40866458Sdfr return (true); 40966458Sdfr if (res0->hr_compression != res1->hr_compression) 41066458Sdfr return (true); 41166458Sdfr if (res0->hr_timeout != res1->hr_timeout) 41266458Sdfr return (true); 41366458Sdfr if (strcmp(res0->hr_exec, res1->hr_exec) != 0) 41466458Sdfr return (true); 41566458Sdfr return (false); 41666458Sdfr} 41766458Sdfr 41866458Sdfrstatic void 41966458Sdfrresource_reload(const struct hast_resource *res) 42066458Sdfr{ 42166458Sdfr struct nv *nvin, *nvout; 42266458Sdfr int error; 42366458Sdfr 42466458Sdfr PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 42566458Sdfr 42666458Sdfr nvout = nv_alloc(); 42766458Sdfr nv_add_uint8(nvout, HASTCTL_RELOAD, "cmd"); 42866458Sdfr nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr"); 42966458Sdfr nv_add_string(nvout, res->hr_sourceaddr, "sourceaddr"); 43066458Sdfr nv_add_int32(nvout, (int32_t)res->hr_replication, "replication"); 43166458Sdfr nv_add_int32(nvout, (int32_t)res->hr_checksum, "checksum"); 43266458Sdfr nv_add_int32(nvout, (int32_t)res->hr_compression, "compression"); 43366458Sdfr nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout"); 43466458Sdfr nv_add_string(nvout, res->hr_exec, "exec"); 43566458Sdfr if (nv_error(nvout) != 0) { 43666458Sdfr nv_free(nvout); 43766458Sdfr pjdlog_error("Unable to allocate header for reload message."); 43866458Sdfr return; 43966458Sdfr } 44066458Sdfr if (hast_proto_send(res, res->hr_ctrl, nvout, NULL, 0) < 0) { 44166458Sdfr pjdlog_errno(LOG_ERR, "Unable to send reload message"); 44266458Sdfr nv_free(nvout); 44366458Sdfr return; 44466458Sdfr } 44566458Sdfr nv_free(nvout); 44666458Sdfr 44766458Sdfr /* Receive response. */ 44866458Sdfr if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) { 44966458Sdfr pjdlog_errno(LOG_ERR, "Unable to receive reload reply"); 45066458Sdfr return; 45166458Sdfr } 45266458Sdfr error = nv_get_int16(nvin, "error"); 45366458Sdfr nv_free(nvin); 45466458Sdfr if (error != 0) { 45566458Sdfr pjdlog_common(LOG_ERR, 0, error, "Reload failed"); 45666458Sdfr return; 45766458Sdfr } 45866458Sdfr} 45966458Sdfr 46066458Sdfrstatic void 46166458Sdfrhastd_reload(void) 46266458Sdfr{ 46366458Sdfr struct hastd_config *newcfg; 46466458Sdfr struct hast_resource *nres, *cres, *tres; 46566458Sdfr uint8_t role; 46666458Sdfr 46766458Sdfr pjdlog_info("Reloading configuration..."); 46866458Sdfr 46966458Sdfr newcfg = yy_config_parse(cfgpath, false); 47066458Sdfr if (newcfg == NULL) 47166458Sdfr goto failed; 47266458Sdfr 47366458Sdfr /* 47466458Sdfr * Check if control address has changed. 475141391Sphk */ 476141391Sphk if (strcmp(cfg->hc_controladdr, newcfg->hc_controladdr) != 0) { 477141391Sphk if (proto_server(newcfg->hc_controladdr, 478141391Sphk &newcfg->hc_controlconn) < 0) { 479141391Sphk pjdlog_errno(LOG_ERR, 480141391Sphk "Unable to listen on control address %s", 481141391Sphk newcfg->hc_controladdr); 482141391Sphk goto failed; 483141391Sphk } 484141391Sphk } 485141391Sphk /* 486141391Sphk * Check if listen address has changed. 487141391Sphk */ 48866458Sdfr if (strcmp(cfg->hc_listenaddr, newcfg->hc_listenaddr) != 0) { 48966458Sdfr if (proto_server(newcfg->hc_listenaddr, 49066458Sdfr &newcfg->hc_listenconn) < 0) { 49166458Sdfr pjdlog_errno(LOG_ERR, "Unable to listen on address %s", 49266458Sdfr newcfg->hc_listenaddr); 49366458Sdfr goto failed; 49466458Sdfr } 49566458Sdfr } 49666458Sdfr /* 49766458Sdfr * Only when both control and listen sockets are successfully 49866458Sdfr * initialized switch them to new configuration. 49966458Sdfr */ 50066458Sdfr if (newcfg->hc_controlconn != NULL) { 50166458Sdfr pjdlog_info("Control socket changed from %s to %s.", 50266458Sdfr cfg->hc_controladdr, newcfg->hc_controladdr); 50366458Sdfr proto_close(cfg->hc_controlconn); 50466458Sdfr cfg->hc_controlconn = newcfg->hc_controlconn; 50566458Sdfr newcfg->hc_controlconn = NULL; 50666458Sdfr strlcpy(cfg->hc_controladdr, newcfg->hc_controladdr, 50766458Sdfr sizeof(cfg->hc_controladdr)); 50866458Sdfr } 509 if (newcfg->hc_listenconn != NULL) { 510 pjdlog_info("Listen socket changed from %s to %s.", 511 cfg->hc_listenaddr, newcfg->hc_listenaddr); 512 proto_close(cfg->hc_listenconn); 513 cfg->hc_listenconn = newcfg->hc_listenconn; 514 newcfg->hc_listenconn = NULL; 515 strlcpy(cfg->hc_listenaddr, newcfg->hc_listenaddr, 516 sizeof(cfg->hc_listenaddr)); 517 } 518 519 /* 520 * Stop and remove resources that were removed from the configuration. 521 */ 522 TAILQ_FOREACH_SAFE(cres, &cfg->hc_resources, hr_next, tres) { 523 TAILQ_FOREACH(nres, &newcfg->hc_resources, hr_next) { 524 if (strcmp(cres->hr_name, nres->hr_name) == 0) 525 break; 526 } 527 if (nres == NULL) { 528 control_set_role(cres, HAST_ROLE_INIT); 529 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 530 pjdlog_info("Resource %s removed.", cres->hr_name); 531 free(cres); 532 } 533 } 534 /* 535 * Move new resources to the current configuration. 536 */ 537 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 538 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 539 if (strcmp(cres->hr_name, nres->hr_name) == 0) 540 break; 541 } 542 if (cres == NULL) { 543 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 544 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 545 pjdlog_info("Resource %s added.", nres->hr_name); 546 } 547 } 548 /* 549 * Deal with modified resources. 550 * Depending on what has changed exactly we might want to perform 551 * different actions. 552 * 553 * We do full resource restart in the following situations: 554 * Resource role is INIT or SECONDARY. 555 * Resource role is PRIMARY and path to local component or provider 556 * name has changed. 557 * In case of PRIMARY, the worker process will be killed and restarted, 558 * which also means removing /dev/hast/<name> provider and 559 * recreating it. 560 * 561 * We do just reload (send SIGHUP to worker process) if we act as 562 * PRIMARY, but only if remote address, replication mode, timeout or 563 * execution path has changed. For those, there is no need to restart 564 * worker process. 565 * If PRIMARY receives SIGHUP, it will reconnect if remote address or 566 * replication mode has changed or simply set new timeout if only 567 * timeout has changed. 568 */ 569 TAILQ_FOREACH_SAFE(nres, &newcfg->hc_resources, hr_next, tres) { 570 TAILQ_FOREACH(cres, &cfg->hc_resources, hr_next) { 571 if (strcmp(cres->hr_name, nres->hr_name) == 0) 572 break; 573 } 574 PJDLOG_ASSERT(cres != NULL); 575 if (resource_needs_restart(cres, nres)) { 576 pjdlog_info("Resource %s configuration was modified, restarting it.", 577 cres->hr_name); 578 role = cres->hr_role; 579 control_set_role(cres, HAST_ROLE_INIT); 580 TAILQ_REMOVE(&cfg->hc_resources, cres, hr_next); 581 free(cres); 582 TAILQ_REMOVE(&newcfg->hc_resources, nres, hr_next); 583 TAILQ_INSERT_TAIL(&cfg->hc_resources, nres, hr_next); 584 control_set_role(nres, role); 585 } else if (resource_needs_reload(cres, nres)) { 586 pjdlog_info("Resource %s configuration was modified, reloading it.", 587 cres->hr_name); 588 strlcpy(cres->hr_remoteaddr, nres->hr_remoteaddr, 589 sizeof(cres->hr_remoteaddr)); 590 strlcpy(cres->hr_sourceaddr, nres->hr_sourceaddr, 591 sizeof(cres->hr_sourceaddr)); 592 cres->hr_replication = nres->hr_replication; 593 cres->hr_checksum = nres->hr_checksum; 594 cres->hr_compression = nres->hr_compression; 595 cres->hr_timeout = nres->hr_timeout; 596 strlcpy(cres->hr_exec, nres->hr_exec, 597 sizeof(cres->hr_exec)); 598 if (cres->hr_workerpid != 0) 599 resource_reload(cres); 600 } 601 } 602 603 yy_config_free(newcfg); 604 pjdlog_info("Configuration reloaded successfully."); 605 return; 606failed: 607 if (newcfg != NULL) { 608 if (newcfg->hc_controlconn != NULL) 609 proto_close(newcfg->hc_controlconn); 610 if (newcfg->hc_listenconn != NULL) 611 proto_close(newcfg->hc_listenconn); 612 yy_config_free(newcfg); 613 } 614 pjdlog_warning("Configuration not reloaded."); 615} 616 617static void 618terminate_workers(void) 619{ 620 struct hast_resource *res; 621 622 pjdlog_info("Termination signal received, exiting."); 623 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 624 if (res->hr_workerpid == 0) 625 continue; 626 pjdlog_info("Terminating worker process (resource=%s, role=%s, pid=%u).", 627 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 628 if (kill(res->hr_workerpid, SIGTERM) == 0) 629 continue; 630 pjdlog_errno(LOG_WARNING, 631 "Unable to send signal to worker process (resource=%s, role=%s, pid=%u).", 632 res->hr_name, role2str(res->hr_role), res->hr_workerpid); 633 } 634} 635 636static void 637listen_accept(void) 638{ 639 struct hast_resource *res; 640 struct proto_conn *conn; 641 struct nv *nvin, *nvout, *nverr; 642 const char *resname; 643 const unsigned char *token; 644 char laddr[256], raddr[256]; 645 size_t size; 646 pid_t pid; 647 int status; 648 649 proto_local_address(cfg->hc_listenconn, laddr, sizeof(laddr)); 650 pjdlog_debug(1, "Accepting connection to %s.", laddr); 651 652 if (proto_accept(cfg->hc_listenconn, &conn) < 0) { 653 pjdlog_errno(LOG_ERR, "Unable to accept connection %s", laddr); 654 return; 655 } 656 657 proto_local_address(conn, laddr, sizeof(laddr)); 658 proto_remote_address(conn, raddr, sizeof(raddr)); 659 pjdlog_info("Connection from %s to %s.", raddr, laddr); 660 661 /* Error in setting timeout is not critical, but why should it fail? */ 662 if (proto_timeout(conn, HAST_TIMEOUT) < 0) 663 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 664 665 nvin = nvout = nverr = NULL; 666 667 /* 668 * Before receiving any data see if remote host have access to any 669 * resource. 670 */ 671 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 672 if (proto_address_match(conn, res->hr_remoteaddr)) 673 break; 674 } 675 if (res == NULL) { 676 pjdlog_error("Client %s isn't known.", raddr); 677 goto close; 678 } 679 /* Ok, remote host can access at least one resource. */ 680 681 if (hast_proto_recv_hdr(conn, &nvin) < 0) { 682 pjdlog_errno(LOG_ERR, "Unable to receive header from %s", 683 raddr); 684 goto close; 685 } 686 687 resname = nv_get_string(nvin, "resource"); 688 if (resname == NULL) { 689 pjdlog_error("No 'resource' field in the header received from %s.", 690 raddr); 691 goto close; 692 } 693 pjdlog_debug(2, "%s: resource=%s", raddr, resname); 694 token = nv_get_uint8_array(nvin, &size, "token"); 695 /* 696 * NULL token means that this is first conection. 697 */ 698 if (token != NULL && size != sizeof(res->hr_token)) { 699 pjdlog_error("Received token of invalid size from %s (expected %zu, got %zu).", 700 raddr, sizeof(res->hr_token), size); 701 goto close; 702 } 703 704 /* 705 * From now on we want to send errors to the remote node. 706 */ 707 nverr = nv_alloc(); 708 709 /* Find resource related to this connection. */ 710 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 711 if (strcmp(resname, res->hr_name) == 0) 712 break; 713 } 714 /* Have we found the resource? */ 715 if (res == NULL) { 716 pjdlog_error("No resource '%s' as requested by %s.", 717 resname, raddr); 718 nv_add_stringf(nverr, "errmsg", "Resource not configured."); 719 goto fail; 720 } 721 722 /* Now that we know resource name setup log prefix. */ 723 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 724 725 /* Does the remote host have access to this resource? */ 726 if (!proto_address_match(conn, res->hr_remoteaddr)) { 727 pjdlog_error("Client %s has no access to the resource.", raddr); 728 nv_add_stringf(nverr, "errmsg", "No access to the resource."); 729 goto fail; 730 } 731 /* Is the resource marked as secondary? */ 732 if (res->hr_role != HAST_ROLE_SECONDARY) { 733 pjdlog_warning("We act as %s for the resource and not as %s as requested by %s.", 734 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY), 735 raddr); 736 nv_add_stringf(nverr, "errmsg", 737 "Remote node acts as %s for the resource and not as %s.", 738 role2str(res->hr_role), role2str(HAST_ROLE_SECONDARY)); 739 if (res->hr_role == HAST_ROLE_PRIMARY) { 740 /* 741 * If we act as primary request the other side to wait 742 * for us a bit, as we might be finishing cleanups. 743 */ 744 nv_add_uint8(nverr, 1, "wait"); 745 } 746 goto fail; 747 } 748 /* Does token (if exists) match? */ 749 if (token != NULL && memcmp(token, res->hr_token, 750 sizeof(res->hr_token)) != 0) { 751 pjdlog_error("Token received from %s doesn't match.", raddr); 752 nv_add_stringf(nverr, "errmsg", "Token doesn't match."); 753 goto fail; 754 } 755 /* 756 * If there is no token, but we have half-open connection 757 * (only remotein) or full connection (worker process is running) 758 * we have to cancel those and accept the new connection. 759 */ 760 if (token == NULL) { 761 PJDLOG_ASSERT(res->hr_remoteout == NULL); 762 pjdlog_debug(1, "Initial connection from %s.", raddr); 763 if (res->hr_workerpid != 0) { 764 PJDLOG_ASSERT(res->hr_remotein == NULL); 765 pjdlog_debug(1, 766 "Worker process exists (pid=%u), stopping it.", 767 (unsigned int)res->hr_workerpid); 768 /* Stop child process. */ 769 if (kill(res->hr_workerpid, SIGINT) < 0) { 770 pjdlog_errno(LOG_ERR, 771 "Unable to stop worker process (pid=%u)", 772 (unsigned int)res->hr_workerpid); 773 /* 774 * Other than logging the problem we 775 * ignore it - nothing smart to do. 776 */ 777 } 778 /* Wait for it to exit. */ 779 else if ((pid = waitpid(res->hr_workerpid, 780 &status, 0)) != res->hr_workerpid) { 781 /* We can only log the problem. */ 782 pjdlog_errno(LOG_ERR, 783 "Waiting for worker process (pid=%u) failed", 784 (unsigned int)res->hr_workerpid); 785 } else { 786 child_exit_log(res->hr_workerpid, status); 787 } 788 child_cleanup(res); 789 } else if (res->hr_remotein != NULL) { 790 char oaddr[256]; 791 792 proto_remote_address(res->hr_remotein, oaddr, 793 sizeof(oaddr)); 794 pjdlog_debug(1, 795 "Canceling half-open connection from %s on connection from %s.", 796 oaddr, raddr); 797 proto_close(res->hr_remotein); 798 res->hr_remotein = NULL; 799 } 800 } 801 802 /* 803 * Checks and cleanups are done. 804 */ 805 806 if (token == NULL) { 807 arc4random_buf(res->hr_token, sizeof(res->hr_token)); 808 nvout = nv_alloc(); 809 nv_add_uint8_array(nvout, res->hr_token, 810 sizeof(res->hr_token), "token"); 811 if (nv_error(nvout) != 0) { 812 pjdlog_common(LOG_ERR, 0, nv_error(nvout), 813 "Unable to prepare return header for %s", raddr); 814 nv_add_stringf(nverr, "errmsg", 815 "Remote node was unable to prepare return header: %s.", 816 strerror(nv_error(nvout))); 817 goto fail; 818 } 819 if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0) { 820 int error = errno; 821 822 pjdlog_errno(LOG_ERR, "Unable to send response to %s", 823 raddr); 824 nv_add_stringf(nverr, "errmsg", 825 "Remote node was unable to send response: %s.", 826 strerror(error)); 827 goto fail; 828 } 829 res->hr_remotein = conn; 830 pjdlog_debug(1, "Incoming connection from %s configured.", 831 raddr); 832 } else { 833 res->hr_remoteout = conn; 834 pjdlog_debug(1, "Outgoing connection to %s configured.", raddr); 835 hastd_secondary(res, nvin); 836 } 837 nv_free(nvin); 838 nv_free(nvout); 839 nv_free(nverr); 840 pjdlog_prefix_set("%s", ""); 841 return; 842fail: 843 if (nv_error(nverr) != 0) { 844 pjdlog_common(LOG_ERR, 0, nv_error(nverr), 845 "Unable to prepare error header for %s", raddr); 846 goto close; 847 } 848 if (hast_proto_send(NULL, conn, nverr, NULL, 0) < 0) { 849 pjdlog_errno(LOG_ERR, "Unable to send error to %s", raddr); 850 goto close; 851 } 852close: 853 if (nvin != NULL) 854 nv_free(nvin); 855 if (nvout != NULL) 856 nv_free(nvout); 857 if (nverr != NULL) 858 nv_free(nverr); 859 proto_close(conn); 860 pjdlog_prefix_set("%s", ""); 861} 862 863static void 864connection_migrate(struct hast_resource *res) 865{ 866 struct proto_conn *conn; 867 int16_t val = 0; 868 869 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 870 871 PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 872 873 if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) { 874 pjdlog_errno(LOG_WARNING, 875 "Unable to receive connection command"); 876 return; 877 } 878 if (proto_client(res->hr_sourceaddr[0] != '\0' ? res->hr_sourceaddr : NULL, 879 res->hr_remoteaddr, &conn) < 0) { 880 val = errno; 881 pjdlog_errno(LOG_WARNING, 882 "Unable to create outgoing connection to %s", 883 res->hr_remoteaddr); 884 goto out; 885 } 886 if (proto_connect(conn, -1) < 0) { 887 val = errno; 888 pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 889 res->hr_remoteaddr); 890 proto_close(conn); 891 goto out; 892 } 893 val = 0; 894out: 895 if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) { 896 pjdlog_errno(LOG_WARNING, 897 "Unable to send reply to connection request"); 898 } 899 if (val == 0 && proto_connection_send(res->hr_conn, conn) < 0) 900 pjdlog_errno(LOG_WARNING, "Unable to send connection"); 901 902 pjdlog_prefix_set("%s", ""); 903} 904 905static void 906check_signals(void) 907{ 908 struct timespec sigtimeout; 909 sigset_t mask; 910 int signo; 911 912 sigtimeout.tv_sec = 0; 913 sigtimeout.tv_nsec = 0; 914 915 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 916 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 917 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 918 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 919 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 920 921 while ((signo = sigtimedwait(&mask, NULL, &sigtimeout)) != -1) { 922 switch (signo) { 923 case SIGINT: 924 case SIGTERM: 925 sigexit_received = true; 926 terminate_workers(); 927 proto_close(cfg->hc_controlconn); 928 exit(EX_OK); 929 break; 930 case SIGCHLD: 931 child_exit(); 932 break; 933 case SIGHUP: 934 hastd_reload(); 935 break; 936 default: 937 PJDLOG_ABORT("Unexpected signal (%d).", signo); 938 } 939 } 940} 941 942static void 943main_loop(void) 944{ 945 struct hast_resource *res; 946 struct timeval seltimeout; 947 int fd, maxfd, ret; 948 time_t lastcheck, now; 949 fd_set rfds; 950 951 lastcheck = time(NULL); 952 seltimeout.tv_sec = REPORT_INTERVAL; 953 seltimeout.tv_usec = 0; 954 955 pjdlog_info("Started successfully, running protocol version %d.", 956 HAST_PROTO_VERSION); 957 958 for (;;) { 959 check_signals(); 960 961 /* Setup descriptors for select(2). */ 962 FD_ZERO(&rfds); 963 maxfd = fd = proto_descriptor(cfg->hc_controlconn); 964 PJDLOG_ASSERT(fd >= 0); 965 FD_SET(fd, &rfds); 966 fd = proto_descriptor(cfg->hc_listenconn); 967 PJDLOG_ASSERT(fd >= 0); 968 FD_SET(fd, &rfds); 969 maxfd = fd > maxfd ? fd : maxfd; 970 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 971 if (res->hr_event == NULL) 972 continue; 973 fd = proto_descriptor(res->hr_event); 974 PJDLOG_ASSERT(fd >= 0); 975 FD_SET(fd, &rfds); 976 maxfd = fd > maxfd ? fd : maxfd; 977 if (res->hr_role == HAST_ROLE_PRIMARY) { 978 /* Only primary workers asks for connections. */ 979 PJDLOG_ASSERT(res->hr_conn != NULL); 980 fd = proto_descriptor(res->hr_conn); 981 PJDLOG_ASSERT(fd >= 0); 982 FD_SET(fd, &rfds); 983 maxfd = fd > maxfd ? fd : maxfd; 984 } else { 985 PJDLOG_ASSERT(res->hr_conn == NULL); 986 } 987 } 988 989 PJDLOG_ASSERT(maxfd + 1 <= (int)FD_SETSIZE); 990 ret = select(maxfd + 1, &rfds, NULL, NULL, &seltimeout); 991 now = time(NULL); 992 if (lastcheck + REPORT_INTERVAL <= now) { 993 hook_check(); 994 lastcheck = now; 995 } 996 if (ret == 0) { 997 /* 998 * select(2) timed out, so there should be no 999 * descriptors to check. 1000 */ 1001 continue; 1002 } else if (ret == -1) { 1003 if (errno == EINTR) 1004 continue; 1005 KEEP_ERRNO((void)pidfile_remove(pfh)); 1006 pjdlog_exit(EX_OSERR, "select() failed"); 1007 } 1008 1009 /* 1010 * Check for signals before we do anything to update our 1011 * info about terminated workers in the meantime. 1012 */ 1013 check_signals(); 1014 1015 if (FD_ISSET(proto_descriptor(cfg->hc_controlconn), &rfds)) 1016 control_handle(cfg); 1017 if (FD_ISSET(proto_descriptor(cfg->hc_listenconn), &rfds)) 1018 listen_accept(); 1019 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) { 1020 if (res->hr_event == NULL) 1021 continue; 1022 if (FD_ISSET(proto_descriptor(res->hr_event), &rfds)) { 1023 if (event_recv(res) == 0) 1024 continue; 1025 /* The worker process exited? */ 1026 proto_close(res->hr_event); 1027 res->hr_event = NULL; 1028 if (res->hr_conn != NULL) { 1029 proto_close(res->hr_conn); 1030 res->hr_conn = NULL; 1031 } 1032 continue; 1033 } 1034 if (res->hr_role == HAST_ROLE_PRIMARY) { 1035 PJDLOG_ASSERT(res->hr_conn != NULL); 1036 if (FD_ISSET(proto_descriptor(res->hr_conn), 1037 &rfds)) { 1038 connection_migrate(res); 1039 } 1040 } else { 1041 PJDLOG_ASSERT(res->hr_conn == NULL); 1042 } 1043 } 1044 } 1045} 1046 1047static void 1048dummy_sighandler(int sig __unused) 1049{ 1050 /* Nothing to do. */ 1051} 1052 1053int 1054main(int argc, char *argv[]) 1055{ 1056 const char *pidfile; 1057 pid_t otherpid; 1058 bool foreground; 1059 int debuglevel; 1060 sigset_t mask; 1061 1062 foreground = false; 1063 debuglevel = 0; 1064 pidfile = HASTD_PIDFILE; 1065 1066 for (;;) { 1067 int ch; 1068 1069 ch = getopt(argc, argv, "c:dFhP:"); 1070 if (ch == -1) 1071 break; 1072 switch (ch) { 1073 case 'c': 1074 cfgpath = optarg; 1075 break; 1076 case 'd': 1077 debuglevel++; 1078 break; 1079 case 'F': 1080 foreground = true; 1081 break; 1082 case 'P': 1083 pidfile = optarg; 1084 break; 1085 case 'h': 1086 default: 1087 usage(); 1088 } 1089 } 1090 argc -= optind; 1091 argv += optind; 1092 1093 pjdlog_init(PJDLOG_MODE_STD); 1094 pjdlog_debug_set(debuglevel); 1095 1096 g_gate_load(); 1097 1098 pfh = pidfile_open(pidfile, 0600, &otherpid); 1099 if (pfh == NULL) { 1100 if (errno == EEXIST) { 1101 pjdlog_exitx(EX_TEMPFAIL, 1102 "Another hastd is already running, pid: %jd.", 1103 (intmax_t)otherpid); 1104 } 1105 /* If we cannot create pidfile from other reasons, only warn. */ 1106 pjdlog_errno(LOG_WARNING, "Unable to open or create pidfile"); 1107 } 1108 1109 cfg = yy_config_parse(cfgpath, true); 1110 PJDLOG_ASSERT(cfg != NULL); 1111 1112 /* 1113 * Restore default actions for interesting signals in case parent 1114 * process (like init(8)) decided to ignore some of them (like SIGHUP). 1115 */ 1116 PJDLOG_VERIFY(signal(SIGHUP, SIG_DFL) != SIG_ERR); 1117 PJDLOG_VERIFY(signal(SIGINT, SIG_DFL) != SIG_ERR); 1118 PJDLOG_VERIFY(signal(SIGTERM, SIG_DFL) != SIG_ERR); 1119 /* 1120 * Because SIGCHLD is ignored by default, setup dummy handler for it, 1121 * so we can mask it. 1122 */ 1123 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 1124 1125 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1126 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1127 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1128 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1129 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1130 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 1131 1132 /* Listen on control address. */ 1133 if (proto_server(cfg->hc_controladdr, &cfg->hc_controlconn) < 0) { 1134 KEEP_ERRNO((void)pidfile_remove(pfh)); 1135 pjdlog_exit(EX_OSERR, "Unable to listen on control address %s", 1136 cfg->hc_controladdr); 1137 } 1138 /* Listen for remote connections. */ 1139 if (proto_server(cfg->hc_listenaddr, &cfg->hc_listenconn) < 0) { 1140 KEEP_ERRNO((void)pidfile_remove(pfh)); 1141 pjdlog_exit(EX_OSERR, "Unable to listen on address %s", 1142 cfg->hc_listenaddr); 1143 } 1144 1145 if (!foreground) { 1146 if (daemon(0, 0) < 0) { 1147 KEEP_ERRNO((void)pidfile_remove(pfh)); 1148 pjdlog_exit(EX_OSERR, "Unable to daemonize"); 1149 } 1150 1151 /* Start logging to syslog. */ 1152 pjdlog_mode_set(PJDLOG_MODE_SYSLOG); 1153 1154 /* Write PID to a file. */ 1155 if (pidfile_write(pfh) < 0) { 1156 pjdlog_errno(LOG_WARNING, 1157 "Unable to write PID to a file"); 1158 } 1159 } 1160 1161 hook_init(); 1162 1163 main_loop(); 1164 1165 exit(0); 1166} 1167